title
Artificial Intelligence Tutorial | Artificial Intelligence Course | Intellipaat

description
🔥Intellipaat Artificial Intelligence Masters Course:- https://intellipaat.com/artificial-intelligence-masters-training-course/ In this Artificial Intelligence tutorial for beginners video you will learn all the major basic concepts in Artificial Intelligence like what is ai, difference between ai, ml and dl, topology of a neural network, how to train network with backpropagation with an in depth demo on tensorflow and keras. #aiartificialintelligence #artificialintelligencecourse #whatisartificialintelligence #artificialintelligenceforbeginners #aitutorialforbeginners 📌 Do subscribe to Intellipaat channel & get regular updates on technological videos: http://bit.ly/Intellipaat 📝This artificial intelligence training video helps you to learn following topics: 00:00 - artificial intelligence tutorial 00:57 - what makes human intelligent? 01:15 - what is artificial intelligence 01:30 - difference between ai, ml and dl 01:45 - why to study artificial intelligence 05:17 - artificial intelligence 07:20 - what is intelligence 08:27 - what makes human intelligent 09:37 - ai vs ml vs dl 11:50 - machine learning applications 17:15 - machine learning types 23:55 - machine learning algorithms 26:11 - what is deep learning 28:33 - deep learning applications 29:40 - how deep learning works 31:00 - what is neural network 31:51 - artificial neural networks 34:30 - artificial neurons 44:50 - deep learning frameworks 🔗 Watch Artificial Intelligence video tutorials here: https://goo.gl/gyf2g3 📕 Read complete Artificial Intelligence tutorial here: https://bit.ly/2nuITZg 📰Interested to learn Artificial Intelligence still more? Please check similar Blogs here:- https://goo.gl/rFFw9L Are you looking for something more? Enroll in our Artificial Intelligence Course and become a certified A.I. professional (https://goo.gl/RdA17B). It is a 32 hrs instructor led AI for everyone training provided by Intellipaat which is completely aligned with industry standards and certification bodies. If you’ve enjoyed this what is ai, ai vs ml video, Like us and Subscribe to our channel for more similar videos and free tutorials. Got any questions? Ask us in the comment section below. ---------------------------- Intellipaat Edge 1. 24*7 Life time Access & Support 2. Flexible Class Schedule 3. Job Assistance 4. Mentors with +14 yrs 5. Industry Oriented Course ware 6. Life time free Course Upgrade ------------------------------ Why should you watch this Artificial Intelligence tutorial? You can learn Artificial Intelligence much faster than any other technology and this Artificial Intelligence tutorial helps you do just that. Artificial Intelligence is one of the best technological advances that is finding increased applications for machine learning and in a lot of industry domains. We are offering the top Artificial Intelligence tutorial that is AI for everyone to gain knowledge in Artificial Intelligence. Our Artificial Intelligence course has been created with extensive inputs from the industry experts so that you can learn Artificial Intelligence and apply it for real world scenarios. Who should watch this Artificial Intelligence tutorial video? If you want to learn Artificial Intelligence to become an A.I. expert then this Intellipaat Artificial Intelligence tutorial and AI deep learning course with tensorflow is for you. The Intellipaat Artificial Intelligence video is your first step to learn A.I. Since this A.I. tutorial and examples video can be taken by anybody, so if you are a beginner in technology then you can also watch other Artificial Intelligence tutorial to take your skills to the next level. Why Artificial Intelligence is important? Artificial Intelligence is taking over each and every industry domain. Machine Learning and especially Deep Learning are the most important aspects of Artificial Intelligence that are being deployed everywhere from search engines to online movie recommendations. Taking the Intellipaat deep learning training & Artificial Intelligence Course can help professionals to build a solid career in a rising technology domain and get the best jobs in top organizations. Why should you opt for a Artificial Intelligence career? If you want to fast-track your career then you should strongly consider Artificial Intelligence. The reason for this is that it is one of the fastest growing technology. There is a huge demand for professionals in Artificial Intelligence. The salaries for A.I. Professionals is fantastic.There is a huge growth opportunity in this domain as well. Hence this Intellipaat Artificial Intelligence tutorial is your stepping stone to a successful career! ------------------------------ For more Information: Please write us to sales@intellipaat.com, or call us at: +91- 7847955955 Website: https://goo.gl/RdA17B Facebook: https://www.facebook.com/intellipaatonline LinkedIn: https://www.linkedin.com/in/intellipaat/ Twitter: https://twitter.com/Intellipaat

detail
{'title': 'Artificial Intelligence Tutorial | Artificial Intelligence Course | Intellipaat', 'heatmap': [{'end': 1067.628, 'start': 635.92, 'weight': 0.736}, {'end': 3193.776, 'start': 1269.758, 'weight': 1}], 'summary': 'The tutorial covers ai basics, impact across industries, machine learning, deep learning, tensorflow basics, neural networks, image processing, mnist, multi-class image classification, and ai implementation challenges. it includes demos with tensorflow and keras, achieving 94% accuracy in imagenet competition, and building models achieving 85% training and 83% testing accuracy.', 'chapters': [{'end': 54.329, 'segs': [{'end': 54.329, 'src': 'embed', 'start': 18.812, 'weight': 0, 'content': [{'end': 25.077, 'text': "we've come up with an end-to-end artificial intelligence session where you learn all the major concepts of artificial intelligence.", 'start': 18.812, 'duration': 6.265}, {'end': 27.819, 'text': 'And before we go ahead and start a session,', 'start': 25.517, 'duration': 2.302}, {'end': 33.023, 'text': 'do subscribe to our channel and like and share our video so that we can create more such informative content.', 'start': 27.819, 'duration': 5.204}, {'end': 34.923, 'text': "So let's have a quick glance at the agenda.", 'start': 33.383, 'duration': 1.54}, {'end': 40.665, 'text': "We'll start off by understanding the difference between artificial intelligence, machine learning, and deep learning.", 'start': 35.364, 'duration': 5.301}, {'end': 43.826, 'text': "And then we'll understand the topology of a neural network.", 'start': 41.125, 'duration': 2.701}, {'end': 47.967, 'text': "Going ahead, we'll understand how can we train our network with backpropagation.", 'start': 44.186, 'duration': 3.781}, {'end': 54.329, 'text': "And finally, we'll implement a demo on the MNIST dataset with TensorFlow and Fashion dataset with Keras.", 'start': 48.327, 'duration': 6.002}], 'summary': 'End-to-end ai session covering major concepts, including ai, ml, and deep learning, with demos on mnist and fashion datasets.', 'duration': 35.517, 'max_score': 18.812, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/W7N6LPp0SmY/pics/W7N6LPp0SmY18812.jpg'}], 'start': 3.24, 'title': 'Ai session: from basics to implementation', 'summary': 'Introduces an end-to-end artificial intelligence session covering major concepts, including ai, machine learning, deep learning, neural network topology, backpropagation, and implementation demos with tensorflow and keras.', 'chapters': [{'end': 54.329, 'start': 3.24, 'title': 'Ai session: from basics to implementation', 'summary': 'Introduces an end-to-end artificial intelligence session covering major concepts, including ai, machine learning, deep learning, neural network topology, backpropagation, and implementation demos with tensorflow and keras.', 'duration': 51.089, 'highlights': ['The session covers major concepts of artificial intelligence, including AI, machine learning, and deep learning.', 'The agenda includes understanding the topology of a neural network and training the network with backpropagation.', 'The session concludes with implementation demos on the MNIST dataset with TensorFlow and Fashion dataset with Keras.', 'Encourages audience engagement by asking them to subscribe, like, and share the content for more informative sessions.']}], 'duration': 51.089, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/W7N6LPp0SmY/pics/W7N6LPp0SmY3240.jpg', 'highlights': ['The session covers major concepts of artificial intelligence, including AI, machine learning, and deep learning.', 'The agenda includes understanding the topology of a neural network and training the network with backpropagation.', 'The session concludes with implementation demos on the MNIST dataset with TensorFlow and Fashion dataset with Keras.', 'Encourages audience engagement by asking them to subscribe, like, and share the content for more informative sessions.']}, {'end': 1161.386, 'segs': [{'end': 89.263, 'src': 'embed', 'start': 55.233, 'weight': 2, 'content': [{'end': 62.081, 'text': 'So tell me, what is it that makes humans intelligent? Well, we as humans can think, learn and make decisions.', 'start': 55.233, 'duration': 6.848}, {'end': 64.063, 'text': 'And that is what makes us intelligent.', 'start': 62.421, 'duration': 1.642}, {'end': 67.848, 'text': 'Now imagine if machines can show human-like intelligence.', 'start': 64.464, 'duration': 3.384}, {'end': 71.432, 'text': 'A machine which can think and make decisions like humans.', 'start': 68.268, 'duration': 3.164}, {'end': 73.394, 'text': "That is truly amazing, isn't it??", 'start': 71.812, 'duration': 1.582}, {'end': 83.7, 'text': 'Artificial intelligence is basically that field of computer science which emphasizes on the creation of intelligent machines which can work and react like humans.', 'start': 74.415, 'duration': 9.285}, {'end': 89.263, 'text': "So now that we know what artificial intelligence is, let's see where do machine learning and deep learning fit in.", 'start': 84.201, 'duration': 5.062}], 'summary': 'Humans are intelligent because they can think, learn, and make decisions; artificial intelligence aims to create machines that can work and react like humans.', 'duration': 34.03, 'max_score': 55.233, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/W7N6LPp0SmY/pics/W7N6LPp0SmY55233.jpg'}, {'end': 164.963, 'src': 'embed', 'start': 138.549, 'weight': 1, 'content': [{'end': 143.393, 'text': 'And at times there is also a chance of a human error in these activities, so to speak.', 'start': 138.549, 'duration': 4.844}, {'end': 154.862, 'text': 'So some of the works that banks and financial institutions handle are investing money in stocks, financial operations, managing various properties,', 'start': 144.114, 'duration': 10.748}, {'end': 155.423, 'text': 'and so on.', 'start': 154.862, 'duration': 0.561}, {'end': 164.963, 'text': 'And with the use of AI system in this process, The institutions are able to achieve efficient results in a quick turnaround time.', 'start': 156.143, 'duration': 8.82}], 'summary': 'Banks and financial institutions use ai to achieve efficient results in investing, managing properties, and financial operations.', 'duration': 26.414, 'max_score': 138.549, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/W7N6LPp0SmY/pics/W7N6LPp0SmY138549.jpg'}, {'end': 240.952, 'src': 'embed', 'start': 189.329, 'weight': 0, 'content': [{'end': 195.91, 'text': 'So there are numerous areas in medical science where AI is used to achieve incredible value.', 'start': 189.329, 'duration': 6.581}, {'end': 202.411, 'text': 'So the help of AI, the medical science was able to create a virtual personal healthcare assistant.', 'start': 196.39, 'duration': 6.021}, {'end': 206.472, 'text': 'So these are used for research and analysis purpose.', 'start': 203.012, 'duration': 3.46}, {'end': 215.594, 'text': 'There are also many efficient healthcare bots introduced in the medical field to provide constant health support to patients.', 'start': 207.072, 'duration': 8.522}, {'end': 219.342, 'text': 'And it is also used in the aerospace industry.', 'start': 216.42, 'duration': 2.922}, {'end': 228.326, 'text': 'So in aerospace, there are a lot of features from booking the tickets to the takeoff and operation of the flights that AI takes care of.', 'start': 219.802, 'duration': 8.524}, {'end': 237.49, 'text': 'AI applications make air transport efficient, fast, safe, and also provides comfortable journey to the passengers.', 'start': 229.006, 'duration': 8.484}, {'end': 240.952, 'text': 'And it has also changed the face of gaming.', 'start': 238.271, 'duration': 2.681}], 'summary': 'Ai in medical science and aerospace improves efficiency, safety, and patient support.', 'duration': 51.623, 'max_score': 189.329, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/W7N6LPp0SmY/pics/W7N6LPp0SmY189329.jpg'}, {'end': 291.671, 'src': 'embed', 'start': 266.848, 'weight': 5, 'content': [{'end': 274.758, 'text': 'Scientists believe that Once the AI system starts working in its full capacity, it will reinvent the world that we know today.', 'start': 266.848, 'duration': 7.91}, {'end': 286.027, 'text': 'So think of the world where all the menial tasks such as garbage disposal, construction, digging and so on will be taken care of by AI applications.', 'start': 275.278, 'duration': 10.749}, {'end': 291.671, 'text': "So it'll all be a time when the hierarchical order dictates the limits of human.", 'start': 286.587, 'duration': 5.084}], 'summary': 'Ai will revolutionize the world, handling menial tasks and redefining human limits.', 'duration': 24.823, 'max_score': 266.848, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/W7N6LPp0SmY/pics/W7N6LPp0SmY266848.jpg'}, {'end': 420.721, 'src': 'embed', 'start': 392.139, 'weight': 4, 'content': [{'end': 400.277, 'text': 'So she is the first humanoid robot who can actually speak to us like natural humans.', 'start': 392.139, 'duration': 8.138}, {'end': 409.039, 'text': 'So Sofia can show some wide range of emotions exhibited by humans, but she is actually a robot.', 'start': 400.397, 'duration': 8.642}, {'end': 413.32, 'text': 'Another application of artificial intelligence is a self-driving car.', 'start': 409.559, 'duration': 3.761}, {'end': 420.721, 'text': 'So you have self-driving cars by Google and Tesla, which actually drive by themselves.', 'start': 413.9, 'duration': 6.821}], 'summary': 'First humanoid robot, sofia, shows human emotions. self-driving cars by google and tesla.', 'duration': 28.582, 'max_score': 392.139, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/W7N6LPp0SmY/pics/W7N6LPp0SmY392139.jpg'}, {'end': 623.461, 'src': 'embed', 'start': 597.114, 'weight': 6, 'content': [{'end': 603.053, 'text': 'Machine learning and deep learning are just ways to achieve artificial intelligence.', 'start': 597.114, 'duration': 5.939}, {'end': 616.218, 'text': 'Now machine learning is that part of artificial intelligence which aims to teach the computers the ability to do tasks with data without any explicit programming.', 'start': 603.533, 'duration': 12.685}, {'end': 623.461, 'text': "Right So we don't need to do any explicit programming and the algorithms do tasks by themselves.", 'start': 616.458, 'duration': 7.003}], 'summary': 'Machine learning teaches computers tasks with data, without explicit programming.', 'duration': 26.347, 'max_score': 597.114, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/W7N6LPp0SmY/pics/W7N6LPp0SmY597114.jpg'}, {'end': 1067.628, 'src': 'heatmap', 'start': 635.92, 'weight': 0.736, 'content': [{'end': 639.903, 'text': 'So first we have AI and then we have ML and then we have DL.', 'start': 635.92, 'duration': 3.983}, {'end': 643.746, 'text': 'So deep learning comes in where machine learning fails.', 'start': 640.224, 'duration': 3.522}, {'end': 651.933, 'text': "And we apply deep learning through something known as artificial neural networks, about which we'll obviously learn later.", 'start': 644.427, 'duration': 7.506}, {'end': 656.817, 'text': "Right So now let's understand artificial intelligence in a bigger set.", 'start': 652.954, 'duration': 3.863}, {'end': 667.406, 'text': "So, as I've already told you, artificial intelligence is the superset under which comes machine learning, under which comes deep learning,", 'start': 657.543, 'duration': 9.863}, {'end': 673.928, 'text': 'and then machine learning and deep learning are basically ways to achieve artificial intelligence.', 'start': 667.406, 'duration': 6.522}, {'end': 678.909, 'text': 'Now, these are the different areas of research of artificial intelligence.', 'start': 674.408, 'duration': 4.501}, {'end': 683.61, 'text': 'So you have ML again, a part of ML is deep learning.', 'start': 679.449, 'duration': 4.161}, {'end': 686.211, 'text': 'Then we have natural language processing.', 'start': 684.09, 'duration': 2.121}, {'end': 698.46, 'text': 'So over here we basically understand what is spoken or written by a human and then we have speech where we either translate the speech to text or we translate the text to speech.', 'start': 686.567, 'duration': 11.893}, {'end': 704.847, 'text': 'The next subfield is robotics and then we have autonomous vehicles under robotics.', 'start': 699.14, 'duration': 5.707}, {'end': 708.871, 'text': 'So Google self-driving car is an example of this over here.', 'start': 705.347, 'duration': 3.524}, {'end': 715.389, 'text': "So, now that we've also understood the difference between artificial intelligence, machine learning and deep learning,", 'start': 709.244, 'duration': 6.145}, {'end': 718.431, 'text': "let's see different examples of machine learning around us.", 'start': 715.389, 'duration': 3.042}, {'end': 721.533, 'text': 'So most of you would have shopped on Amazon.', 'start': 718.911, 'duration': 2.622}, {'end': 727.478, 'text': 'Now, when you go into Amazon, you see that there are some products recommended to you.', 'start': 721.913, 'duration': 5.565}, {'end': 729.479, 'text': 'Now, how do you think that would happen?', 'start': 727.818, 'duration': 1.661}, {'end': 738.586, 'text': 'So this is something known as a recommendation engine, and recommendation engine is nothing but a component of machine learning.', 'start': 730.08, 'duration': 8.506}, {'end': 744.294, 'text': "So let's say you and your friend buy similar products.", 'start': 739.133, 'duration': 5.161}, {'end': 749.015, 'text': 'So your friend buys five products and you buy three products.', 'start': 744.714, 'duration': 4.301}, {'end': 755.836, 'text': 'Now out of those, whatever three products you buy are same as what your friend buys.', 'start': 749.475, 'duration': 6.361}, {'end': 764.738, 'text': "So let's say the common products are an iPhone, a back cover for the iPhone and a Bluetooth headset.", 'start': 756.356, 'duration': 8.382}, {'end': 770.088, 'text': "Now, let's say the other two products bought by a friend would be a MacBook and a mouse.", 'start': 765.305, 'duration': 4.783}, {'end': 774.891, 'text': 'Now, since there are three products which are same between you two.', 'start': 770.428, 'duration': 4.463}, {'end': 781.416, 'text': 'this is why the products which your friend has also bought, those are the products which will be recommended to you.', 'start': 774.891, 'duration': 6.525}, {'end': 790.822, 'text': 'So on the basis of the commonality between you and your friend, you will be recommended a MacBook and a mouse as well.', 'start': 781.816, 'duration': 9.006}, {'end': 794.144, 'text': 'So this is nothing but a concept of machine learning.', 'start': 791.222, 'duration': 2.922}, {'end': 796.977, 'text': 'And then we have Amazon Alexa.', 'start': 795.137, 'duration': 1.84}, {'end': 802.759, 'text': 'So Amazon Alexa is, so Amazon Alexa is a really good example of speech recognition.', 'start': 797.417, 'duration': 5.342}, {'end': 806.359, 'text': "You know, when you say Alexa, turn on the lights, it'll turn on the lights.", 'start': 803.059, 'duration': 3.3}, {'end': 810.34, 'text': "When you say Alexa, book a ride for me, it'll do exactly that.", 'start': 806.719, 'duration': 3.621}, {'end': 816.741, 'text': 'When you say Alexa, order a cheese pizza, and that is exactly what Amazon Alexa will do.', 'start': 810.76, 'duration': 5.981}, {'end': 819.522, 'text': 'Now, Alexa is just a machine, right?', 'start': 817.061, 'duration': 2.461}, {'end': 828.411, 'text': 'But When you say do something, order a pizza, book a cab for me, turn on the lights, you know how is the machine able to understand all of this?', 'start': 819.942, 'duration': 8.469}, {'end': 835.072, 'text': 'So the idea behind this is speech recognition, and that is again a component of machine learning.', 'start': 829.011, 'duration': 6.061}, {'end': 838.032, 'text': "And then we have Netflix's movie recommendation.", 'start': 835.412, 'duration': 2.62}, {'end': 840.733, 'text': "So let's say you watch two TV series.", 'start': 838.412, 'duration': 2.321}, {'end': 845.994, 'text': 'First TV series is Friends and the next TV series is Big Bang Theory.', 'start': 841.033, 'duration': 4.961}, {'end': 856.349, 'text': 'and Since you watch these two TV series which belong to the genre comedy, that is why maybe you will be recommended How I Met Your Mother,', 'start': 845.994, 'duration': 10.355}, {'end': 863.257, 'text': 'or you can be recommended Silicon Valley or some other TV series belonging to the comedy genre.', 'start': 856.349, 'duration': 6.908}, {'end': 865.44, 'text': 'So this again is machine learning.', 'start': 863.638, 'duration': 1.802}, {'end': 868.353, 'text': 'And then we also have Google traffic prediction.', 'start': 865.911, 'duration': 2.442}, {'end': 877.56, 'text': "Let's just say you're traveling in your car and there is huge traffic there and you desperately want to get out of the traffic.", 'start': 868.733, 'duration': 8.827}, {'end': 886.286, 'text': 'So you turn on Google maps and Google maps tells you the best direction from where the traffic would be the least.', 'start': 877.94, 'duration': 8.346}, {'end': 890.409, 'text': 'Now, how does Google maps do this? This again is machine learning.', 'start': 886.847, 'duration': 3.562}, {'end': 898.966, 'text': "So now that we've looked at different real world applications of machine learning, Let's actually understand what exactly is machine learning.", 'start': 890.589, 'duration': 8.377}, {'end': 902.047, 'text': "So, as I've already told you,", 'start': 899.626, 'duration': 2.421}, {'end': 911.991, 'text': 'machine learning is a subset of artificial intelligence which gives the machine ability to learn without being explicitly programmed.', 'start': 902.047, 'duration': 9.944}, {'end': 923.236, 'text': 'So over here, data is the key, or in other words, you basically teach a machine how to learn without any explicit programming.', 'start': 912.011, 'duration': 11.225}, {'end': 927.238, 'text': 'and the machine learns with the help of data.', 'start': 923.757, 'duration': 3.481}, {'end': 931.579, 'text': "So let's just go through this simple video to understand more about machine learning.", 'start': 927.418, 'duration': 4.161}, {'end': 969.84, 'text': "Right So now that we know what exactly is machine learning, let's also understand how does machine learning work.", 'start': 962.018, 'duration': 7.822}, {'end': 975.381, 'text': "So as I've already told you, machine learning depends totally on data.", 'start': 970.5, 'duration': 4.881}, {'end': 980.802, 'text': 'So first, we take in a data set and divide it into two parts.', 'start': 975.801, 'duration': 5.001}, {'end': 983.843, 'text': 'The first part would be the training set.', 'start': 981.242, 'duration': 2.601}, {'end': 986.883, 'text': 'And the second part would be the testing set.', 'start': 984.283, 'duration': 2.6}, {'end': 992.097, 'text': 'And we will train the model on top of the training site.', 'start': 987.223, 'duration': 4.874}, {'end': 1003.723, 'text': 'So now once we train the model we will give it new data and check for its accuracy on top of that new data.', 'start': 992.538, 'duration': 11.185}, {'end': 1013.708, 'text': 'And if the accuracy of that new data comes out to be good enough then we will go ahead and use that machine learning model.', 'start': 1004.283, 'duration': 9.425}, {'end': 1016.909, 'text': 'On the other hand, the model which you build.', 'start': 1014.288, 'duration': 2.621}, {'end': 1025.891, 'text': "if the accuracy of that model is not good enough, then We'll go ahead and fine tune that model till we get the desired accuracy.", 'start': 1016.909, 'duration': 8.982}, {'end': 1029.973, 'text': 'This is the basic premise behind machine learning.', 'start': 1026.55, 'duration': 3.423}, {'end': 1033.137, 'text': "Now let's look at the subcategories of machine learning.", 'start': 1030.374, 'duration': 2.763}, {'end': 1037.88, 'text': 'So we have supervised learning, unsupervised learning and reinforcement learning.', 'start': 1033.577, 'duration': 4.303}, {'end': 1042.864, 'text': 'So in supervised learning, you can consider that the learning is guided by a teacher.', 'start': 1038.181, 'duration': 4.683}, {'end': 1050.19, 'text': 'So we have a data set which actually acts as a teacher and its role is to train the model or the machine.', 'start': 1043.484, 'duration': 6.706}, {'end': 1058.225, 'text': 'So once the model gets trained it can start making a prediction or decision when new data is given to it.', 'start': 1050.782, 'duration': 7.443}, {'end': 1060.065, 'text': "So let's take this example.", 'start': 1058.785, 'duration': 1.28}, {'end': 1067.628, 'text': 'So over here we are training this machine by giving it samples of data.', 'start': 1060.625, 'duration': 7.003}], 'summary': 'Ai encompasses ml, dl, nlp, and robotics. ml powers amazon recommendations and netflix movie suggestions. machine learning depends on data and involves supervised, unsupervised, and reinforcement learning.', 'duration': 431.708, 'max_score': 635.92, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/W7N6LPp0SmY/pics/W7N6LPp0SmY635920.jpg'}, {'end': 738.586, 'src': 'embed', 'start': 709.244, 'weight': 7, 'content': [{'end': 715.389, 'text': "So, now that we've also understood the difference between artificial intelligence, machine learning and deep learning,", 'start': 709.244, 'duration': 6.145}, {'end': 718.431, 'text': "let's see different examples of machine learning around us.", 'start': 715.389, 'duration': 3.042}, {'end': 721.533, 'text': 'So most of you would have shopped on Amazon.', 'start': 718.911, 'duration': 2.622}, {'end': 727.478, 'text': 'Now, when you go into Amazon, you see that there are some products recommended to you.', 'start': 721.913, 'duration': 5.565}, {'end': 729.479, 'text': 'Now, how do you think that would happen?', 'start': 727.818, 'duration': 1.661}, {'end': 738.586, 'text': 'So this is something known as a recommendation engine, and recommendation engine is nothing but a component of machine learning.', 'start': 730.08, 'duration': 8.506}], 'summary': "Examples of machine learning like amazon's recommendation engine explained.", 'duration': 29.342, 'max_score': 709.244, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/W7N6LPp0SmY/pics/W7N6LPp0SmY709244.jpg'}], 'start': 55.233, 'title': "Ai's impact across industries", 'summary': 'Discusses the omnipresence and impact of ai across industries, including banking, medical science, aerospace, and gaming, highlighting its role in enhancing efficiency, accuracy, and customer service while also envisioning its potential to revolutionize various tasks in the future.', 'chapters': [{'end': 111.157, 'start': 55.233, 'title': 'Understanding artificial intelligence', 'summary': 'Discusses the concept of artificial intelligence, emphasizing on how it enables machines to exhibit human-like intelligence, and its relevance in various domains.', 'duration': 55.924, 'highlights': ['Artificial intelligence enables machines to exhibit human-like intelligence, including the abilities to think, learn, and make decisions, which is truly amazing.', 'Machine learning and deep learning are subsets of artificial intelligence, providing means to achieve human-like intelligence in machines.', 'Artificial intelligence is pervasive and has relevance in various domains, making it essential to study and understand its implications.']}, {'end': 420.721, 'start': 111.157, 'title': "Ai's impact across industries", 'summary': 'Discusses the omnipresence and impact of ai across industries, including banking, medical science, aerospace, and gaming, highlighting its role in enhancing efficiency, accuracy, and customer service while also envisioning its potential to revolutionize various tasks in the future.', 'duration': 309.564, 'highlights': ["AI's applications in banking and finance enable efficient results and quick resolution, reducing time and effort from employees. Efficient results, quick resolution, reduced time and effort", "AI's wide applications in medical science have led to the creation of virtual personal healthcare assistants and efficient healthcare bots, revolutionizing healthcare support for patients. Virtual personal healthcare assistants, efficient healthcare bots, revolutionized healthcare support", 'In aerospace, AI applications make air transport efficient, fast, safe, and provide a comfortable journey to passengers. Efficient air transport, safety, comfortable journey', 'AI has revolutionized gaming, enabling a whole new level of gameplay with its applications. Revolutionized gaming, enhanced gameplay', 'Envisioning the potential of AI to revolutionize various tasks in the future, such as garbage disposal, construction, and more, leading to a world where human limitations and hierarchical orders are transcended. Revolutionizing various tasks, transcending human limitations and hierarchical orders', "AI's definition and existing applications, including chatbots like OK Google and Siri, and humanoid robots like Sophia, along with self-driving cars by Google and Tesla, demonstrate its wide-ranging impact and potential. Wide-ranging impact, existing applications, AI's potential"]}, {'end': 1161.386, 'start': 420.861, 'title': 'Understanding artificial intelligence and machine learning', 'summary': 'Explores the concept of artificial intelligence, its subsets machine learning and deep learning, and real-world applications of machine learning such as recommendation engines, speech recognition, and traffic prediction.', 'duration': 740.525, 'highlights': ['Artificial Intelligence defined as capacity for understanding, self-awareness, learning, and problem solving. Intelligence is defined as capacity for understanding, self-awareness, learning, and problem solving, which can be applied to machines as artificial intelligence.', 'Machine learning aims to teach computers tasks without explicit programming, using numerical and statistical approaches. Machine learning aims to teach computers tasks without explicit programming, using numerical and statistical approaches to achieve artificial intelligence.', "Real-world applications of machine learning include recommendation engines, speech recognition, and traffic prediction. Real-world applications of machine learning include recommendation engines, speech recognition, and traffic prediction, exemplified by Amazon's recommendation engine, Amazon Alexa's speech recognition, Netflix's movie recommendation, and Google traffic prediction."]}], 'duration': 1106.153, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/W7N6LPp0SmY/pics/W7N6LPp0SmY55233.jpg', 'highlights': ["AI's wide applications in medical science have led to the creation of virtual personal healthcare assistants and efficient healthcare bots, revolutionizing healthcare support for patients.", "AI's applications in banking and finance enable efficient results and quick resolution, reducing time and effort from employees.", 'Artificial intelligence enables machines to exhibit human-like intelligence, including the abilities to think, learn, and make decisions, which is truly amazing.', "AI's applications in aerospace make air transport efficient, fast, safe, and provide a comfortable journey to passengers.", "AI's definition and existing applications, including chatbots like OK Google and Siri, and humanoid robots like Sophia, along with self-driving cars by Google and Tesla, demonstrate its wide-ranging impact and potential.", 'Envisioning the potential of AI to revolutionize various tasks in the future, such as garbage disposal, construction, and more, leading to a world where human limitations and hierarchical orders are transcended.', 'Machine learning aims to teach computers tasks without explicit programming, using numerical and statistical approaches to achieve artificial intelligence.', "Real-world applications of machine learning include recommendation engines, speech recognition, and traffic prediction, exemplified by Amazon's recommendation engine, Amazon Alexa's speech recognition, Netflix's movie recommendation, and Google traffic prediction.", 'Artificial intelligence is pervasive and has relevance in various domains, making it essential to study and understand its implications.']}, {'end': 2892.992, 'segs': [{'end': 1194.051, 'src': 'embed', 'start': 1161.746, 'weight': 0, 'content': [{'end': 1166.089, 'text': 'So this spam classification is basically an example of supervised learning.', 'start': 1161.746, 'duration': 4.343}, {'end': 1169.965, 'text': 'Then we have unsupervised learning.', 'start': 1168.024, 'duration': 1.941}, {'end': 1176.186, 'text': 'So in unsupervised learning, the model learns through observation and finds structures in the data.', 'start': 1170.325, 'duration': 5.861}, {'end': 1185.669, 'text': 'So once the model is given a data set, it automatically finds patterns and relationships in the data set by creating clusters in it.', 'start': 1176.766, 'duration': 8.903}, {'end': 1194.051, 'text': 'So what it cannot do is add labels to the cluster, like it cannot say this is a group of apples or mangoes,', 'start': 1186.169, 'duration': 7.882}], 'summary': 'Supervised and unsupervised learning explained with data structure examples.', 'duration': 32.305, 'max_score': 1161.746, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/W7N6LPp0SmY/pics/W7N6LPp0SmY1161746.jpg'}, {'end': 1294.804, 'src': 'embed', 'start': 1262.793, 'weight': 2, 'content': [{'end': 1266.095, 'text': 'But when compared with these three objects, they are very dissimilar.', 'start': 1262.793, 'duration': 3.302}, {'end': 1269.277, 'text': 'This is the underlying concept of unsupervised learning.', 'start': 1266.275, 'duration': 3.002}, {'end': 1275.621, 'text': 'And a good example of unsupervised learning would be again Netflix movie recommendation.', 'start': 1269.758, 'duration': 5.863}, {'end': 1282.726, 'text': 'So over here, the movies are segregated on the basis of different genres.', 'start': 1276.062, 'duration': 6.664}, {'end': 1294.804, 'text': 'So over here, TV series like Friends, How I Met Your Mother and Silicon Valley are clustered into one group because those come into the same category.', 'start': 1283.106, 'duration': 11.698}], 'summary': 'Unsupervised learning clusters similar items, like netflix recommends movies based on genres.', 'duration': 32.011, 'max_score': 1262.793, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/W7N6LPp0SmY/pics/W7N6LPp0SmY1262793.jpg'}, {'end': 1600.011, 'src': 'embed', 'start': 1552.453, 'weight': 3, 'content': [{'end': 1557.454, 'text': "So this is the lack of creativity that I'm talking about when it comes to machine learning.", 'start': 1552.453, 'duration': 5.001}, {'end': 1564.384, 'text': 'And also there are a lot of time constraints as the model has to learn through a lot of historical data.', 'start': 1557.861, 'duration': 6.523}, {'end': 1567.346, 'text': 'So that was everything about machine learning.', 'start': 1564.765, 'duration': 2.581}, {'end': 1569.967, 'text': "Now let's start off with deep learning.", 'start': 1567.566, 'duration': 2.401}, {'end': 1581.894, 'text': 'So deep learning is a subset of machine learning where it learns through data representations as opposed to task specific algorithms.', 'start': 1570.508, 'duration': 11.386}, {'end': 1591.124, 'text': 'So we saw that the drawback in machine learning models was that the models are specific to only one particular task.', 'start': 1582.534, 'duration': 8.59}, {'end': 1600.011, 'text': 'But this is not the case with deep learning models, as these deep learning models are based on the data representations.', 'start': 1591.564, 'duration': 8.447}], 'summary': 'Machine learning lacks creativity, deep learning uses data representations and is not task-specific.', 'duration': 47.558, 'max_score': 1552.453, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/W7N6LPp0SmY/pics/W7N6LPp0SmY1552453.jpg'}, {'end': 1848.652, 'src': 'embed', 'start': 1796.035, 'weight': 5, 'content': [{'end': 1804.642, 'text': 'So a deep neural network basically has these three models an input layer, the hidden layers and the output layer.', 'start': 1796.035, 'duration': 8.607}, {'end': 1809.606, 'text': 'And the term deep usually refers to the number of hidden layers in the neural network.', 'start': 1805.162, 'duration': 4.444}, {'end': 1819.686, 'text': 'So traditionally neural networks only contain two to three hidden layers while deep networks can have as many as 150 hidden layers.', 'start': 1810.246, 'duration': 9.44}, {'end': 1822.468, 'text': "Now that's a very huge amount, isn't it?", 'start': 1820.186, 'duration': 2.282}, {'end': 1837.176, 'text': 'So deep learning models are trained by using large sets of label data and neural network architectures that learn features directly from the data without the need for manual feature extraction.', 'start': 1823.028, 'duration': 14.148}, {'end': 1848.652, 'text': 'So all of the input data is given to this input layer and this input layer automatically extracts the features by itself.', 'start': 1837.777, 'duration': 10.875}], 'summary': 'Deep neural networks can have up to 150 hidden layers, automatically extracting features from input data.', 'duration': 52.617, 'max_score': 1796.035, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/W7N6LPp0SmY/pics/W7N6LPp0SmY1796035.jpg'}, {'end': 2282.762, 'src': 'embed', 'start': 2258.874, 'weight': 7, 'content': [{'end': 2267.437, 'text': 'So with the third dimension, I have introduced non-linearity in our data which helps in creating a linearly separable model.', 'start': 2258.874, 'duration': 8.563}, {'end': 2272.079, 'text': "And in real-world situations, you don't always get linear problems.", 'start': 2268.197, 'duration': 3.882}, {'end': 2275.46, 'text': 'So you should know how to deal with non-linear problems as well.', 'start': 2272.519, 'duration': 2.941}, {'end': 2282.762, 'text': 'And this is where activation functions help us to convert the linear equation to non-linear form.', 'start': 2275.92, 'duration': 6.842}], 'summary': 'Introducing non-linearity in data for creating a linearly separable model and converting linear equations to non-linear form using activation functions.', 'duration': 23.888, 'max_score': 2258.874, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/W7N6LPp0SmY/pics/W7N6LPp0SmY2258874.jpg'}, {'end': 2331.647, 'src': 'embed', 'start': 2306.995, 'weight': 8, 'content': [{'end': 2312.577, 'text': "Now there are many types of activation functions and today we'll be discussing some of the widely used ones.", 'start': 2306.995, 'duration': 5.582}, {'end': 2315.6, 'text': "So let's start with the identity function.", 'start': 2313.519, 'duration': 2.081}, {'end': 2320.462, 'text': 'So the identity function gives out the same output as the input.', 'start': 2316.16, 'duration': 4.302}, {'end': 2326.305, 'text': 'So, no matter how many layers we have, if all the activations are identity functions,', 'start': 2320.902, 'duration': 5.403}, {'end': 2331.647, 'text': 'then the final output of the last layer would be the same as the input given to the first layer.', 'start': 2326.305, 'duration': 5.342}], 'summary': 'Discusses various activation functions, starting with the identity function and its impact on output.', 'duration': 24.652, 'max_score': 2306.995, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/W7N6LPp0SmY/pics/W7N6LPp0SmY2306995.jpg'}, {'end': 2501.441, 'src': 'embed', 'start': 2445.805, 'weight': 9, 'content': [{'end': 2452.794, 'text': 'And finally, we have the softmax function, which is ideally used in the output layer for classification problems.', 'start': 2445.805, 'duration': 6.989}, {'end': 2461.172, 'text': 'So the softmax function basically gives a set of probability values for each class of the output,', 'start': 2453.806, 'duration': 7.366}, {'end': 2466.616, 'text': 'and that particular class which would have the maximum probability will be our output class.', 'start': 2461.172, 'duration': 5.444}, {'end': 2470.019, 'text': 'So that was all about activation functions.', 'start': 2467.917, 'duration': 2.102}, {'end': 2472.501, 'text': 'Now let us learn more about perceptrons.', 'start': 2470.299, 'duration': 2.202}, {'end': 2479.461, 'text': 'So like we were taught how to behave in certain conditions, perceptrons also require training.', 'start': 2473.595, 'duration': 5.866}, {'end': 2483.446, 'text': 'So they have a learning algorithm through which they produce the output.', 'start': 2480.242, 'duration': 3.204}, {'end': 2495.539, 'text': 'By training a perceptron we try to find a line plane or some hyperplane which can accurately separate these two classes by adjusting the weights and biases.', 'start': 2483.986, 'duration': 11.553}, {'end': 2501.441, 'text': 'So consider this image where we give the dogs and horses as input.', 'start': 2496.536, 'duration': 4.905}], 'summary': 'Softmax function assigns probability values to output classes for classification, while perceptrons require training to find a separating hyperplane for class separation.', 'duration': 55.636, 'max_score': 2445.805, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/W7N6LPp0SmY/pics/W7N6LPp0SmY2445805.jpg'}, {'end': 2633.325, 'src': 'embed', 'start': 2603.561, 'weight': 11, 'content': [{'end': 2607.522, 'text': 'So artificial neural networks have the ability to generalize their inputs.', 'start': 2603.561, 'duration': 3.961}, {'end': 2612.803, 'text': 'This ability is valuable for robotics and pattern recognition systems.', 'start': 2608.122, 'duration': 4.681}, {'end': 2617.238, 'text': 'Artificial neural networks also help in nonlinear data processing.', 'start': 2613.792, 'duration': 3.446}, {'end': 2624.169, 'text': 'So nonlinear systems have the capability of finding shortcuts to reach computationally expensive solutions.', 'start': 2617.899, 'duration': 6.27}, {'end': 2633.325, 'text': 'These systems can also infer connections between data points rather than waiting for records in a data source to be explicitly linked.', 'start': 2624.878, 'duration': 8.447}], 'summary': 'Artificial neural networks excel in generalizing inputs, aiding robotics and pattern recognition, and processing nonlinear data efficiently.', 'duration': 29.764, 'max_score': 2603.561, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/W7N6LPp0SmY/pics/W7N6LPp0SmY2603561.jpg'}, {'end': 2720.445, 'src': 'embed', 'start': 2691.331, 'weight': 12, 'content': [{'end': 2696.913, 'text': 'Now to implement these artificial neural networks, you would need the help of a deep learning framework.', 'start': 2691.331, 'duration': 5.582}, {'end': 2704.417, 'text': "So the first question to pop into your head would be, what are the different deep learning frameworks available? So today we'll cover just that.", 'start': 2697.494, 'duration': 6.923}, {'end': 2706.218, 'text': "So let's start with TensorFlow.", 'start': 2704.997, 'duration': 1.221}, {'end': 2711.32, 'text': 'So TensorFlow is arguably one of the best deep learning frameworks that we have today.', 'start': 2706.718, 'duration': 4.602}, {'end': 2720.445, 'text': 'It is an open source software library developed by the researchers and engineers from the Google Brain team for high performance numerical computation.', 'start': 2711.9, 'duration': 8.545}], 'summary': 'Tensorflow is a leading deep learning framework developed by google brain team.', 'duration': 29.114, 'max_score': 2691.331, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/W7N6LPp0SmY/pics/W7N6LPp0SmY2691331.jpg'}], 'start': 1161.746, 'title': 'Machine learning and deep learning', 'summary': 'Explains unsupervised learning, types of machine learning, including reinforcement learning and its use in self-driving cars, applications of deep learning in speech recognition, self-driving cars, and language translation, and activation functions in neural networks and deep learning frameworks, including tensorflow, keras, pytorch, dl4j, and mxnet.', 'chapters': [{'end': 1303.531, 'start': 1161.746, 'title': 'Unsupervised learning in data segregation', 'summary': 'Explains unsupervised learning, where the model identifies patterns and relationships in data by creating clusters, illustrated through a fruit segregation example and netflix movie recommendation, showcasing its ability to find similarities without predefined labels.', 'duration': 141.785, 'highlights': ['Unsupervised learning involves the model learning through observation and finding structures in the data.', 'The model automatically finds patterns and relationships in the data set by creating clusters in it.', 'The unsupervised learning model segregates fruits based on similar characteristics into distinct clusters.', 'The underlying concept of unsupervised learning is to find similarities and dissimilarities in data without predefined labels.', 'Netflix movie recommendation is a good example of unsupervised learning, where movies are segregated based on different genres and lead actors.']}, {'end': 1703.472, 'start': 1304.111, 'title': 'Types of machine learning & limitations', 'summary': 'Explains the three types of machine learning: supervised, unsupervised, and reinforcement learning, with a focus on reinforcement learning and its use in self-driving cars. it also touches upon the limitations of machine learning, including the need for massive training data, difficulty in error diagnosis, lack of creativity, and time constraints.', 'duration': 399.361, 'highlights': ['The chapter explains the three types of machine learning: supervised, unsupervised, and reinforcement learning It discusses the concept of reinforcement learning, where an agent interacts with an environment, observes the outcomes of its actions, and adjusts its behavior based on rewards and penalties. It also mentions the use of reinforcement learning in self-driving cars.', 'The limitations of machine learning are discussed, including the need for massive training data, difficulty in error diagnosis, lack of creativity, and time constraints It emphasizes the requirement for large training data for accurate results, the challenge of error diagnosis due to the complexity of algorithms, the lack of versatility in machine learning models, and the time constraints in learning from historical data.', 'Deep learning is introduced as a subset of machine learning that learns through data representations and utilizes deep neural networks It contrasts deep learning with traditional machine learning by highlighting the automatic feature extraction in deep learning models, as opposed to manual feature extraction in machine learning. It also mentions that deep learning performance improves with more data, in contrast to machine learning.']}, {'end': 2210.956, 'start': 1703.472, 'title': 'Applications and working of deep learning', 'summary': 'Discusses the applications of deep learning, including speech recognition, self-driving cars, and language translation, and explains the working of deep learning through neural network architectures, hidden layers, and the training process using large sets of labeled data.', 'duration': 507.484, 'highlights': ['Deep learning models are often referred to as deep neural networks with as many as 150 hidden layers, allowing for automatic feature extraction from input data. Deep neural networks can have as many as 150 hidden layers, enabling automatic feature extraction from input data.', 'Neural network architectures in deep learning learn features directly from the data without the need for manual feature extraction, and are trained using large sets of labeled data. Neural network architectures in deep learning learn features directly from the data without the need for manual feature extraction and are trained using large sets of labeled data.', "The input layer, hidden layers, and the output layer in a deep neural network structure resemble the network structure of neurons in the brain, and the term 'deep' refers to the number of hidden layers, which can range up to 150. The deep neural network structure resembles the network structure of neurons in the brain, with the term 'deep' referring to the number of hidden layers, which can range up to 150.", 'The artificial neural network, inspired by the biological neural network of the brain, processes input data, performs calculations, and uses the output to solve problems, similar to the biological neural network consisting of dendrites, cell body, and axon. The artificial neural network processes input data, performs calculations, and uses the output to solve problems, similar to the biological neural network consisting of dendrites, cell body, and axon.', 'The artificial neuron, called a perceptron, takes an input, processes it, passes it through an activation function, and returns the output if the condition is met, serving as the fundamental unit of deep neural networks. The artificial neuron, called a perceptron, serves as the fundamental unit of deep neural networks by taking an input, processing it, passing it through an activation function, and returning the output if the condition is met.']}, {'end': 2444.484, 'start': 2212.036, 'title': 'Activation functions in neural networks', 'summary': 'Discusses the need for activation functions to handle non-linear problems, introduces different types of activation functions, and explains their functions, ranges, and applications in neural networks.', 'duration': 232.448, 'highlights': ['Activation functions are necessary to handle non-linear problems and create linearly separable models. By introducing a third dimension, non-linearity is added to the equation, making it easier to create a linearly separable model.', 'Different types of widely used activation functions are discussed, including identity function, binary step function, sigmoid function, tanh function, ReLU function, and leaky ReLU. The chapter explains the functions, ranges, and applications of the identity function, binary step function, sigmoid function, tanh function, ReLU function, and leaky ReLU in neural networks.', 'ReLU function is the most widely used activation function and is primarily implemented on the hidden layers of the neural network. The ReLU function acts as an identity function for input greater than or equal to zero and is widely used on the hidden layers of neural networks.']}, {'end': 2892.992, 'start': 2445.805, 'title': 'Activation functions and deep learning frameworks', 'summary': 'Covers the softmax function for classification problems, perceptron training algorithm, benefits of artificial neural networks, and overview of deep learning frameworks, including tensorflow, keras, pytorch, dl4j, and mxnet.', 'duration': 447.187, 'highlights': ['The softmax function gives probability values for each class of the output, with the class having the maximum probability being the output class. The softmax function is used in the output layer for classification problems, providing probability values for each class, and determining the class with the maximum probability as the output.', 'Perceptrons require training through a learning algorithm to produce output by adjusting weights and biases. Perceptrons require training to find a hyperplane that accurately separates classes by adjusting weights and biases, with training iterations reducing error values for classifying inputs.', 'Artificial neural networks have benefits including organic learning, input generalization, nonlinear data processing, fault tolerance, and self-repairing capabilities. Artificial neural networks offer organic learning, input generalization, nonlinear data processing, fault tolerance, and self-repairing capabilities, making them valuable in various applications.', 'TensorFlow, Keras, PyTorch, DL4J, and MXNet are prominent deep learning frameworks with unique features and use cases. The chapter provides an overview of TensorFlow, Keras, PyTorch, DL4J, and MXNet as prominent deep learning frameworks, highlighting their unique features and use cases.']}], 'duration': 1731.246, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/W7N6LPp0SmY/pics/W7N6LPp0SmY1161746.jpg', 'highlights': ['Unsupervised learning involves model learning through observation and finding structures in the data.', 'The model automatically finds patterns and relationships in the data set by creating clusters in it.', 'The underlying concept of unsupervised learning is to find similarities and dissimilarities in data without predefined labels.', 'The limitations of machine learning are discussed, including the need for massive training data, difficulty in error diagnosis, lack of creativity, and time constraints.', 'Deep learning is introduced as a subset of machine learning that learns through data representations and utilizes deep neural networks.', 'Neural network architectures in deep learning learn features directly from the data without the need for manual feature extraction and are trained using large sets of labeled data.', 'The input layer, hidden layers, and the output layer in a deep neural network structure resemble the network structure of neurons in the brain.', 'Activation functions are necessary to handle non-linear problems and create linearly separable models.', 'Different types of widely used activation functions are discussed, including identity function, binary step function, sigmoid function, tanh function, ReLU function, and leaky ReLU.', 'The softmax function gives probability values for each class of the output, with the class having the maximum probability being the output class.', 'Perceptrons require training through a learning algorithm to produce output by adjusting weights and biases.', 'Artificial neural networks have benefits including organic learning, input generalization, nonlinear data processing, fault tolerance, and self-repairing capabilities.', 'TensorFlow, Keras, PyTorch, DL4J, and MXNet are prominent deep learning frameworks with unique features and use cases.']}, {'end': 4213.431, 'segs': [{'end': 2946.405, 'src': 'embed', 'start': 2918.686, 'weight': 0, 'content': [{'end': 2922.748, 'text': 'So you can consider these tensors to be the building blocks in TensorFlow.', 'start': 2918.686, 'duration': 4.062}, {'end': 2927.41, 'text': 'Now, these very tensors are given as the input for the neural network.', 'start': 2923.388, 'duration': 4.022}, {'end': 2932.168, 'text': "So as I've said, a tensor is nothing but an n dimensional array.", 'start': 2928.243, 'duration': 3.925}, {'end': 2936.673, 'text': 'So the number of dimensions used to represent the data is known as its rank.', 'start': 2932.748, 'duration': 3.925}, {'end': 2946.405, 'text': 'So if a tensor has just one element, In other words, if it has just magnitude and no direction, then its rank will be 0.', 'start': 2937.454, 'duration': 8.951}], 'summary': "Tensors are building blocks in tensorflow, used as input for neural networks. a tensor's rank is determined by the number of dimensions, with a rank of 0 for a single-element tensor.", 'duration': 27.719, 'max_score': 2918.686, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/W7N6LPp0SmY/pics/W7N6LPp0SmY2918686.jpg'}, {'end': 2988.522, 'src': 'embed', 'start': 2959.369, 'weight': 1, 'content': [{'end': 2964.671, 'text': 'Now TensorFlow as the name states is a combination of two words, tensor and flow.', 'start': 2959.369, 'duration': 5.302}, {'end': 2970.954, 'text': 'Here the data is stored in tensors, but the execution is done in the form of a graph.', 'start': 2965.292, 'duration': 5.662}, {'end': 2978.758, 'text': 'So this is not like your traditional programming where you just write a bunch of lines and everything gets executed in sequence.', 'start': 2971.734, 'duration': 7.024}, {'end': 2988.522, 'text': "So first you'd have to prepare this computational graph and then this computational graph is executed inside something known as a session.", 'start': 2979.358, 'duration': 9.164}], 'summary': 'Tensorflow stores data in tensors and executes operations in a computational graph, not in traditional sequential programming.', 'duration': 29.153, 'max_score': 2959.369, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/W7N6LPp0SmY/pics/W7N6LPp0SmY2959369.jpg'}, {'end': 3111.096, 'src': 'embed', 'start': 3083.562, 'weight': 3, 'content': [{'end': 3086.464, 'text': 'Now inside this I will give the value of the constant.', 'start': 3083.562, 'duration': 2.902}, {'end': 3089.567, 'text': "So let's say the value is 10.", 'start': 3087.065, 'duration': 2.502}, {'end': 3091.869, 'text': 'So this is an integer type constant.', 'start': 3089.567, 'duration': 2.302}, {'end': 3098.611, 'text': "Now, similarly, I'll also create a floating type constant and I'll store this in CON2.", 'start': 3092.589, 'duration': 6.022}, {'end': 3107.414, 'text': "So I'll type tf.constant and the floating value would be 3.14.", 'start': 3099.311, 'duration': 8.103}, {'end': 3111.096, 'text': "Now, after this, I'll create a string type constant.", 'start': 3107.414, 'duration': 3.682}], 'summary': 'Creating integer and floating type constants with values 10 and 3.14, and a string type constant.', 'duration': 27.534, 'max_score': 3083.562, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/W7N6LPp0SmY/pics/W7N6LPp0SmY3083562.jpg'}, {'end': 3477.192, 'src': 'embed', 'start': 3443.873, 'weight': 4, 'content': [{'end': 3448.416, 'text': 'And again, 5 cross 1 gives us a 5.', 'start': 3443.873, 'duration': 4.543}, {'end': 3452.278, 'text': 'So this was addition and multiplication with respect to lists.', 'start': 3448.416, 'duration': 3.862}, {'end': 3457.961, 'text': 'Now let me also do a simple operation on strings.', 'start': 3455.3, 'duration': 2.661}, {'end': 3462.163, 'text': 'So let me take in the first string and name it as str1.', 'start': 3458.762, 'duration': 3.401}, {'end': 3463.944, 'text': 'So this is a constant.', 'start': 3462.944, 'duration': 1}, {'end': 3472.409, 'text': "So tf.constant and let's say I type over here I love and then I give a space.", 'start': 3464.165, 'duration': 8.244}, {'end': 3477.192, 'text': 'Now I will take in the second string which would be str2.', 'start': 3473.43, 'duration': 3.762}], 'summary': 'Demonstrated addition and multiplication with lists and simple string operations in tensorflow.', 'duration': 33.319, 'max_score': 3443.873, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/W7N6LPp0SmY/pics/W7N6LPp0SmY3443873.jpg'}, {'end': 3631.215, 'src': 'embed', 'start': 3602.095, 'weight': 5, 'content': [{'end': 3606.099, 'text': "So this is when I'll feed the value to this placeholder A over here.", 'start': 3602.095, 'duration': 4.004}, {'end': 3610.183, 'text': 'Now to do that, I would have to create something known as a feed dictionary.', 'start': 3606.739, 'duration': 3.444}, {'end': 3616.067, 'text': 'So feed dict equals, let me create a dictionary over here.', 'start': 3611.365, 'duration': 4.702}, {'end': 3622.811, 'text': "So it would be A and the value which I'll be giving to A would be, let's say five.", 'start': 3616.868, 'duration': 5.943}, {'end': 3631.215, 'text': "Now let me run this and let's see what do we get, right? So during the execution time, I have assigned a value of five to A.", 'start': 3623.511, 'duration': 7.704}], 'summary': 'Creating a feed dictionary to assign a value of 5 to placeholder a.', 'duration': 29.12, 'max_score': 3602.095, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/W7N6LPp0SmY/pics/W7N6LPp0SmY3602095.jpg'}, {'end': 3953.028, 'src': 'embed', 'start': 3923.85, 'weight': 6, 'content': [{'end': 3929.855, 'text': "So let's say I assign this variable a value of 20 and this is of integer type.", 'start': 3923.85, 'duration': 6.005}, {'end': 3932.797, 'text': 'So tf.int32.', 'start': 3929.935, 'duration': 2.862}, {'end': 3934.318, 'text': "I'll run this.", 'start': 3933.658, 'duration': 0.66}, {'end': 3942.826, 'text': 'Now another thing to be kept in mind is whenever we are declaring values in TensorFlow, they have to be initialized.', 'start': 3935.119, 'duration': 7.707}, {'end': 3945.568, 'text': 'So this is how we can initialize all of the variables.', 'start': 3943.186, 'duration': 2.382}, {'end': 3953.028, 'text': 'So we have something known as global variable initializer.', 'start': 3947.546, 'duration': 5.482}], 'summary': 'Assign variable a value of 20 of integer type in tensorflow and initialize variables using global variable initializer.', 'duration': 29.178, 'max_score': 3923.85, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/W7N6LPp0SmY/pics/W7N6LPp0SmY3923850.jpg'}, {'end': 4013, 'src': 'embed', 'start': 3983.281, 'weight': 7, 'content': [{'end': 3988.286, 'text': 'Now since this is a variable the value of a variable can be actually updated.', 'start': 3983.281, 'duration': 5.005}, {'end': 3991.63, 'text': 'So let me go ahead and update the value of this.', 'start': 3988.787, 'duration': 2.843}, {'end': 4003.256, 'text': 'So I will name this as updated var 1 and the function would be tf.assign.', 'start': 3993.152, 'duration': 10.104}, {'end': 4008.518, 'text': "And inside this, the first parameter would be the variable which I'd want to update.", 'start': 4004.536, 'duration': 3.982}, {'end': 4013, 'text': "And after that, I need to give the value to which I'd want to update this.", 'start': 4009.198, 'duration': 3.802}], 'summary': "Updating variable 'var 1' using tf.assign function.", 'duration': 29.719, 'max_score': 3983.281, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/W7N6LPp0SmY/pics/W7N6LPp0SmY3983281.jpg'}, {'end': 4098.52, 'src': 'embed', 'start': 4071.42, 'weight': 8, 'content': [{'end': 4078.282, 'text': 'so initially the value of var1 was 20, but we have updated it and made its value to be 25.', 'start': 4071.42, 'duration': 6.862}, {'end': 4081.403, 'text': "now let's also go ahead and create a small linear model.", 'start': 4078.282, 'duration': 3.121}, {'end': 4085.801, 'text': 'So let me just type in linear model over here.', 'start': 4082.776, 'duration': 3.025}, {'end': 4088.344, 'text': 'And this is how our linear model would look like.', 'start': 4086.061, 'duration': 2.283}, {'end': 4095.836, 'text': 'W, X plus B, where W and B would be variables and X would be a placeholder.', 'start': 4088.364, 'duration': 7.472}, {'end': 4098.52, 'text': 'Right, right.', 'start': 4096.737, 'duration': 1.783}], 'summary': 'Initial var1 value was 20, updated to 25. created a linear model.', 'duration': 27.1, 'max_score': 4071.42, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/W7N6LPp0SmY/pics/W7N6LPp0SmY4071420.jpg'}], 'start': 2893.813, 'title': 'Understanding tensorflow basics', 'summary': 'Covers the concept of tensors, tensorflow basics, operations, sessions, placeholders, and variables, with examples up to rank 2, creation of constants, execution within a session, creation and feeding values to placeholders, and creation of variables for trainable parameters.', 'chapters': [{'end': 2958.389, 'start': 2893.813, 'title': 'Tensorflow: understanding tensors and neural networks', 'summary': 'Explains the concept of tensors in tensorflow, highlighting their role as the building blocks for neural networks and defining their rank based on the number of dimensions, with examples up to rank 2.', 'duration': 64.576, 'highlights': ['Tensors are the building blocks in TensorFlow, representing data in the form of multi-dimensional arrays. Tensors in TensorFlow serve as the foundational elements for representing data, covering various use cases such as imaging, speech recognition, forecasting, and NLP.', 'Tensors are used as input for the neural network, with their rank determined by the number of dimensions. The role of tensors as inputs for neural networks is essential, and their rank, indicating the number of dimensions, is crucial for understanding their structure and usage.', 'The rank of a tensor determines the number of dimensions used to represent the data. The concept of rank in tensors is defined based on the number of dimensions used to represent the data, providing a clear understanding of their structure and complexity.']}, {'end': 3171.44, 'start': 2959.369, 'title': 'Tensorflow basics', 'summary': 'Introduces tensorflow, explaining its data storage in tensors and execution in a computational graph, and demonstrates the creation of constants in tensorflow with integer, float, string, and boolean values.', 'duration': 212.071, 'highlights': ['The chapter introduces TensorFlow, explaining its data storage in tensors and execution in a computational graph. TensorFlow stores data in tensors and executes it in a computational graph, depicting mathematical operations as nodes and tensors as edges.', 'Demonstrates the creation of constants in TensorFlow with integer, float, string, and boolean values. The transcript demonstrates the creation of constants in TensorFlow with integer, float, string, and boolean values, and prints their respective tensor types.']}, {'end': 3443.873, 'start': 3172, 'title': 'Tensorflow operations and sessions', 'summary': 'Demonstrates the creation of constants, execution within a session, and performing basic operations like addition and multiplication on tensorflow constants, highlighting the results and higher dimension tensors.', 'duration': 271.873, 'highlights': ["The value of constant 3 is 'this is part of', and the value of constant 4 is false.", 'The value of constant 2 is 3.14 and constant 1 is 10.', 'When we multiply 2 with 4, we get 8.']}, {'end': 3674.725, 'start': 3443.873, 'title': 'Tensorflow operations and placeholders', 'summary': 'Covers addition and multiplication operations on lists and strings using tensorflow, creating and feeding values to placeholders during runtime, with examples of integer and string placeholders resulting in quantifiable data such as 10 and arrays of values.', 'duration': 230.852, 'highlights': ['The chapter covers addition and multiplication operations on lists and strings using TensorFlow Demonstrates addition and multiplication operations on lists and strings using TensorFlow', 'creating and feeding values to placeholders during runtime Illustrates the creation of placeholders and feeding values during runtime', 'examples of integer and string placeholders resulting in quantifiable data such as 10 and arrays of values Shows examples of integer and string placeholders resulting in quantifiable data such as 10 and arrays of values']}, {'end': 4213.431, 'start': 3677.066, 'title': 'Tensorflow basics: placeholders and variables', 'summary': 'Covers the creation and usage of placeholders and variables in tensorflow, including defining placeholders for strings, creating variables for trainable parameters, and updating variable values, as well as creating a linear model with variables and placeholders.', 'duration': 536.365, 'highlights': ["Creating placeholders and assigning values during execution time The chapter demonstrates the creation of placeholders for strings and assigning values during execution time using tf.placeholder, with examples of assigning string values such as 'Sam', 'Bob', and 'Charlie' during runtime.", 'Defining and initializing variables in TensorFlow The process of defining variables in TensorFlow using tf.variable, initializing them with specific values and data types, and ensuring all variables are initialized using global variable initializer is explained, with an example of initializing a variable var1 with a value of 20.', 'Updating variable values with tf.assign The concept of updating variable values using tf.assign is demonstrated, with an example of updating the value of a variable from 20 to 25, showcasing the ability to modify variable values in TensorFlow.', 'Creating a linear model with variables and placeholders The creation of a linear model in TensorFlow, involving the use of variables for W and B, as well as a placeholder for X, and the execution of the linear model within a session after initializing the variables is illustrated.']}], 'duration': 1319.618, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/W7N6LPp0SmY/pics/W7N6LPp0SmY2893813.jpg', 'highlights': ['Tensors are the building blocks in TensorFlow, representing data in the form of multi-dimensional arrays.', 'The chapter introduces TensorFlow, explaining its data storage in tensors and execution in a computational graph.', 'The rank of a tensor determines the number of dimensions used to represent the data.', 'Demonstrates the creation of constants in TensorFlow with integer, float, string, and boolean values.', 'The chapter covers addition and multiplication operations on lists and strings using TensorFlow.', 'Illustrates the creation of placeholders and feeding values during runtime.', 'Defining and initializing variables in TensorFlow.', 'Updating variable values with tf.assign.', 'Creating a linear model with variables and placeholders.']}, {'end': 5277.266, 'segs': [{'end': 4359.067, 'src': 'embed', 'start': 4299.58, 'weight': 0, 'content': [{'end': 4308.263, 'text': 'But after 20 years, almost in like late 90s, 1999, he again came up with this paper.', 'start': 4299.58, 'duration': 8.683}, {'end': 4312.224, 'text': 'He brought in his paper, but this time with a proof.', 'start': 4309.383, 'duration': 2.841}, {'end': 4317.884, 'text': 'And he won a contest that is called ImageNet in 2004.', 'start': 4312.905, 'duration': 4.979}, {'end': 4320.545, 'text': 'This time he came back and everybody had to listen.', 'start': 4317.884, 'duration': 2.661}, {'end': 4324.487, 'text': 'So we were at 2004 when he won the ImageNet.', 'start': 4321.246, 'duration': 3.241}, {'end': 4332.391, 'text': 'I have two people who I have already told that, you know, ImageNet is.', 'start': 4324.727, 'duration': 7.664}, {'end': 4334.772, 'text': 'ImageNet is actually in a competition.', 'start': 4332.391, 'duration': 2.381}, {'end': 4342.076, 'text': 'It is held by a joint division of Stanford plus Princeton University.', 'start': 4335.052, 'duration': 7.024}, {'end': 4348.36, 'text': 'They these guys come up and they they did they downloaded a corpus of images now.', 'start': 4343.017, 'duration': 5.343}, {'end': 4354.764, 'text': 'They have a very large collection of images around 1 million images around covering thousand categories.', 'start': 4348.4, 'duration': 6.364}, {'end': 4359.067, 'text': 'They have you can just click on explore and you can see this tree around here.', 'start': 4355.425, 'duration': 3.642}], 'summary': 'In 1999, he presented a paper with a proof, and won the 2004 imagenet contest, which featured a collection of 1 million images across thousand categories.', 'duration': 59.487, 'max_score': 4299.58, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/W7N6LPp0SmY/pics/W7N6LPp0SmY4299580.jpg'}, {'end': 4426.077, 'src': 'embed', 'start': 4401.024, 'weight': 1, 'content': [{'end': 4407.806, 'text': 'right, it might be a top view, a side view, a bottom view, any view might come up, and at that time people used to fail.', 'start': 4401.024, 'duration': 6.782}, {'end': 4416.728, 'text': 'but in 2014, 2004, when you know, uh, joffrey hilton came with it with his paper, his paper, the competition he won was with 94 accuracy,', 'start': 4407.806, 'duration': 8.922}, {'end': 4422.049, 'text': 'and that was never achieved, never achieved, and it surpassed the human accuracy also.', 'start': 4416.728, 'duration': 5.321}, {'end': 4423.615, 'text': 'so that was the.', 'start': 4422.049, 'duration': 1.566}, {'end': 4426.077, 'text': 'that was where you know, when people started listening to him.', 'start': 4423.615, 'duration': 2.462}], 'summary': 'In 2004, joffrey hilton achieved 94% accuracy, surpassing human accuracy, leading to recognition.', 'duration': 25.053, 'max_score': 4401.024, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/W7N6LPp0SmY/pics/W7N6LPp0SmY4401024.jpg'}, {'end': 4468.665, 'src': 'embed', 'start': 4441.487, 'weight': 3, 'content': [{'end': 4452.394, 'text': 'what joffrey said about um neural network is that he said that that artificial neural network is a computational model that is inspired by the,', 'start': 4441.487, 'duration': 10.907}, {'end': 4455.882, 'text': 'by the way, biological neural networks in our brain.', 'start': 4452.941, 'duration': 2.941}, {'end': 4458.042, 'text': 'human information possesses.', 'start': 4455.882, 'duration': 2.16}, {'end': 4465.805, 'text': 'right. so, um, i mean most of you know this thing, that you know that our nervous system is, is made up of exams.', 'start': 4458.042, 'duration': 7.763}, {'end': 4468.665, 'text': "and uh, okay, i'll show you the photo that you know.", 'start': 4465.805, 'duration': 2.86}], 'summary': 'Joffrey discussed artificial neural networks, inspired by biological neural networks, in our brain.', 'duration': 27.178, 'max_score': 4441.487, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/W7N6LPp0SmY/pics/W7N6LPp0SmY4441487.jpg'}, {'end': 5036.727, 'src': 'embed', 'start': 5007.504, 'weight': 4, 'content': [{'end': 5010.125, 'text': "It's the smallest part of a neural network that you can pick,", 'start': 5007.504, 'duration': 2.621}, {'end': 5016.827, 'text': 'and the smallest part of a neural network is doing this like it takes all the inputs that are coming to itself, give them some weightage,', 'start': 5010.125, 'duration': 6.702}, {'end': 5023.45, 'text': 'multiply the weightage to the corresponding input and then sum all these and then add a small bias to it also.', 'start': 5016.827, 'duration': 6.623}, {'end': 5024.61, 'text': "That's it over.", 'start': 5023.69, 'duration': 0.92}, {'end': 5032.186, 'text': 'How is this weightage and bias being introduced that took Who is okay x1 and x naught x2 and x1.', 'start': 5025.231, 'duration': 6.955}, {'end': 5036.727, 'text': 'I can say okay, someone is giving to us the input that what is the temperature of the glass?', 'start': 5032.206, 'duration': 4.521}], 'summary': "Neural network's smallest part multiplies inputs, sums them, and adds bias.", 'duration': 29.223, 'max_score': 5007.504, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/W7N6LPp0SmY/pics/W7N6LPp0SmY5007504.jpg'}, {'end': 5269.584, 'src': 'embed', 'start': 5240.453, 'weight': 5, 'content': [{'end': 5243.255, 'text': 'So we will be adjusting the marks based upon the grade.', 'start': 5240.453, 'duration': 2.802}, {'end': 5245.897, 'text': 'This is a very hard-coded problem.', 'start': 5244.175, 'duration': 1.722}, {'end': 5256.405, 'text': "I think but let me take an example that you're working on a machine learning problem where you are trying to find out the loan defaults right that in bank.", 'start': 5245.957, 'duration': 10.448}, {'end': 5258.366, 'text': 'You have multiple people who take the loan.', 'start': 5256.485, 'duration': 1.881}, {'end': 5262.061, 'text': 'now the bank wants to know will this person repay the loan or not?', 'start': 5258.366, 'duration': 3.695}, {'end': 5265.762, 'text': 'right now you already have a historical data with you.', 'start': 5262.061, 'duration': 3.701}, {'end': 5267.363, 'text': 'the bank will give the data to you.', 'start': 5265.762, 'duration': 1.601}, {'end': 5269.584, 'text': "okay, let's say that this is the data with you.", 'start': 5267.363, 'duration': 2.221}], 'summary': 'Adjusting marks based on grades, addressing a hard-coded problem using a machine learning example in banking to predict loan defaults.', 'duration': 29.131, 'max_score': 5240.453, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/W7N6LPp0SmY/pics/W7N6LPp0SmY5240453.jpg'}], 'start': 4213.431, 'title': 'Neural networks', 'summary': "Discusses the implementation of a neural network in the imagenet competition, achieving 94% accuracy, and the comparison between artificial neural networks and the human brain's structure and processing speed, along with the basic artificial neuron structure and neural network basics.", 'chapters': [{'end': 4401.024, 'start': 4213.431, 'title': 'Implementation of neural network and imagenet competition', 'summary': "Discusses the implementation of a neural network with a placeholder x and the success story of duffy hilton's idea of mimicking the human mind and winning the imagenet competition in 2004, organized by stanford and princeton university with a dataset of 1 million images covering 1000 categories.", 'duration': 187.593, 'highlights': ["Duffy Hilton's success in the ImageNet competition in 2004 Duffy Hilton proposed the idea of mimicking the human mind and won the ImageNet competition in 2004 with a dataset of 1 million images covering 1000 categories.", 'Proposal and proof of mimicking the human mind by Duffy Hilton Duffy Hilton proposed the idea of mimicking the human mind, initially facing skepticism, but later won the ImageNet competition in 2004 with a dataset of 1 million images covering 1000 categories.', 'Description of ImageNet competition by Stanford and Princeton University The ImageNet competition was organized by Stanford and Princeton University with a dataset of 1 million images covering 1000 categories, aiming to build a model for image classification.', 'Challenges of hand-coded classifiers in image classification Hand-coded classifiers faced challenges in image classification as they could not adapt to the various ways in which objects appear, leading to the need for a more dynamic approach.', "Initial skepticism towards Duffy Hilton's idea of mimicking the human mind Initially, Duffy Hilton's idea of mimicking the human mind faced skepticism and criticism from others in the field, but later, he proved his concept by winning the ImageNet competition in 2004."]}, {'end': 4734.922, 'start': 4401.024, 'title': 'Neural network and human brain', 'summary': "Discusses the breakthrough in neural network accuracy achieved in 2004, reaching 94% and surpassing human accuracy, and provides an in-depth comparison between artificial neural networks and the human brain's neural network structure and processing speed.", 'duration': 333.898, 'highlights': ["The breakthrough in neural network accuracy reached 94% in 2004, surpassing human accuracy. In 2004, Joffrey Hilton's paper described a neural network achieving 94% accuracy, surpassing human accuracy.", "Comparison between artificial neural networks and the human brain's neural network structure and processing speed. The discussion provides a detailed comparison between the structure and processing speed of artificial neural networks and the human brain's neural network.", "Explanation of the flow of information in the human nervous system and its connection to the brain's decision-making process. The transcript explains the flow of information in the human nervous system and its connection to the brain's decision-making process, highlighting the brain's role as the controller of body actions."]}, {'end': 5068.807, 'start': 4734.922, 'title': 'Basic artificial neuron structure', 'summary': 'Discusses the structure and function of a basic artificial neuron, emphasizing its role in processing and transmitting information, and the introduction of weightage and bias in neural networks.', 'duration': 333.885, 'highlights': ['A basic artificial neuron collects input, assigns weightage to each input, multiplies the weightage to the corresponding input, sums them, and adds a bias to produce an output, serving as the smallest part of a neural network. N/A', "The basic neuron's role in transmitting information is emphasized, explaining that the activation of the neuron is crucial for sending information to the brain, demonstrating its significance in processing and transmitting sensory information. N/A", 'The introduction of weightage and bias in artificial neural networks is discussed, highlighting that at the very first instance, they are randomly allocated, providing insight into the initial allocation process in neural network environments. N/A']}, {'end': 5277.266, 'start': 5068.807, 'title': 'Neural network basics', 'summary': 'Explains the basics of a neural network, including the concept of neurons, weights, and inputs, and provides examples of its application in solving problems like grading students based on marks and predicting loan defaults in banks with historical data.', 'duration': 208.459, 'highlights': ['Neural network involves adjusting weights to obtain the correct output, illustrated through examples like grading students based on marks and predicting loan defaults in banks with historical data. The process of adjusting weights in the neural network is demonstrated through examples like grading students based on marks and predicting loan defaults in banks with historical data.', 'Neural network basics include the concept of neurons, weights, and inputs, along with the application of a function to the weighted sum of inputs to compute the output. The basics of neural network, including neurons, weights, and inputs, and the application of a function to the weighted sum of inputs to compute the output.', 'The chapter emphasizes the role of neural networks in solving complex problems, such as predicting loan defaults in banks using historical data. The chapter emphasizes the role of neural networks in solving complex problems, such as predicting loan defaults in banks using historical data.']}], 'duration': 1063.835, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/W7N6LPp0SmY/pics/W7N6LPp0SmY4213431.jpg', 'highlights': ["Duffy Hilton's success in the ImageNet competition in 2004 with a dataset of 1 million images covering 1000 categories.", 'The breakthrough in neural network accuracy reached 94% in 2004, surpassing human accuracy.', 'The ImageNet competition was organized by Stanford and Princeton University with a dataset of 1 million images covering 1000 categories, aiming to build a model for image classification.', "The discussion provides a detailed comparison between the structure and processing speed of artificial neural networks and the human brain's neural network.", 'A basic artificial neuron collects input, assigns weightage to each input, multiplies the weightage to the corresponding input, sums them, and adds a bias to produce an output, serving as the smallest part of a neural network.', 'The process of adjusting weights in the neural network is demonstrated through examples like grading students based on marks and predicting loan defaults in banks with historical data.', 'The basics of neural network, including neurons, weights, and inputs, and the application of a function to the weighted sum of inputs to compute the output.', 'The chapter emphasizes the role of neural networks in solving complex problems, such as predicting loan defaults in banks using historical data.']}, {'end': 7621.626, 'segs': [{'end': 5324.86, 'src': 'embed', 'start': 5296.783, 'weight': 3, 'content': [{'end': 5304.705, 'text': "fine, so the f function that you see this is called the affine, affine thing, and it's doing nothing right.", 'start': 5296.783, 'duration': 7.922}, {'end': 5306.866, 'text': 'you saw that this is it doing a linear thing.', 'start': 5304.705, 'duration': 2.161}, {'end': 5313.689, 'text': "here we're just multiplying the weights, the corresponding x1, x2, and you might now consider x1, x2 as the importance feature.", 'start': 5306.866, 'duration': 6.823}, {'end': 5318.791, 'text': 'so the more important x1 is the of, the more higher the value of w1 will be right,', 'start': 5313.689, 'duration': 5.102}, {'end': 5324.86, 'text': "the more Lower the value of W2 is it means it doesn't make much sense to Y?", 'start': 5318.791, 'duration': 6.069}], 'summary': 'The affine function performs linear operations on weights and features, with higher importance leading to higher weight values.', 'duration': 28.077, 'max_score': 5296.783, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/W7N6LPp0SmY/pics/W7N6LPp0SmY5296783.jpg'}, {'end': 6067.898, 'src': 'embed', 'start': 6035.44, 'weight': 4, 'content': [{'end': 6038.082, 'text': 'what does it expects?', 'start': 6035.44, 'duration': 2.642}, {'end': 6042.906, 'text': 'it should be connected to all the inputs that are in the previous layer.', 'start': 6038.082, 'duration': 4.824}, {'end': 6047.01, 'text': "so this is, let's say, this is called layer input layer.", 'start': 6042.906, 'duration': 4.104}, {'end': 6050.206, 'text': 'Wherever your hidden neurons.', 'start': 6048.885, 'duration': 1.321}, {'end': 6051.427, 'text': 'this is called hidden layer.', 'start': 6050.206, 'duration': 1.221}, {'end': 6055.73, 'text': 'And your outputs are here.', 'start': 6054.449, 'duration': 1.281}, {'end': 6058.432, 'text': 'This is called a output.', 'start': 6056.09, 'duration': 2.342}, {'end': 6067.898, 'text': "Now when you are building the speed forward neural network your responsibility is just one that whichever neuron you're standing.", 'start': 6059.853, 'duration': 8.045}], 'summary': 'Describes the structure of a neural network with input, hidden, and output layers.', 'duration': 32.458, 'max_score': 6035.44, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/W7N6LPp0SmY/pics/W7N6LPp0SmY6035440.jpg'}, {'end': 6425.679, 'src': 'embed', 'start': 6399.374, 'weight': 1, 'content': [{'end': 6408.559, 'text': 'The only thing that you have to do is that you have to come up with the optimal number of neurons and optimal number of layers for each neural network.', 'start': 6399.374, 'duration': 9.185}, {'end': 6411.78, 'text': 'And now who is going to decide this very good question.', 'start': 6409.479, 'duration': 2.301}, {'end': 6415.222, 'text': 'This is you who is going to decide this and how okay.', 'start': 6412.06, 'duration': 3.162}, {'end': 6425.679, 'text': "So first answer is that hit-and-try that everybody will tell you and I'm also telling you that the first approach is hit-and-try.", 'start': 6416.192, 'duration': 9.487}], 'summary': 'Optimize neural network with hit-and-try approach.', 'duration': 26.305, 'max_score': 6399.374, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/W7N6LPp0SmY/pics/W7N6LPp0SmY6399374.jpg'}, {'end': 6885.83, 'src': 'embed', 'start': 6862.321, 'weight': 2, 'content': [{'end': 6870.744, 'text': 'So what Suresh is asking is that know that I took two words in overfitting and underfitting, and how do we address this problem?', 'start': 6862.321, 'duration': 8.423}, {'end': 6877.147, 'text': 'So what we do is that we divide our data into two parts, and I think you must have also done this in ml, also right?', 'start': 6871.505, 'duration': 5.642}, {'end': 6880.768, 'text': 'Training data and validation data right?', 'start': 6879.067, 'duration': 1.701}, {'end': 6885.83, 'text': 'Not to just three parts like training, validation and testing right?', 'start': 6881.848, 'duration': 3.982}], 'summary': 'Suresh asks about overfitting and underfitting, and how to address it by dividing data into training and validation sets.', 'duration': 23.509, 'max_score': 6862.321, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/W7N6LPp0SmY/pics/W7N6LPp0SmY6862321.jpg'}, {'end': 7566.723, 'src': 'embed', 'start': 7524.793, 'weight': 0, 'content': [{'end': 7534.697, 'text': 'so to reduce this error, you go back and check your values of W and switch the values of W to such an extent so that they minimize this error.', 'start': 7524.793, 'duration': 9.904}, {'end': 7550.082, 'text': 'right, whatever I said just now, statistically right, make sense to everyone, right what I just said here?', 'start': 7534.697, 'duration': 15.385}, {'end': 7564.102, 'text': 'okay, now this thing is called back propagation, that now this thing is called back propagation.', 'start': 7550.082, 'duration': 14.02}, {'end': 7566.723, 'text': 'now back propagation, if I like.', 'start': 7564.102, 'duration': 2.621}], 'summary': 'To reduce error, adjust values of w to minimize it. this is called back propagation.', 'duration': 41.93, 'max_score': 7524.793, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/W7N6LPp0SmY/pics/W7N6LPp0SmY7524793.jpg'}], 'start': 5277.266, 'title': 'Neural network basics', 'summary': 'Explains the concept of neural networks, including input layers, hidden layers, and backpropagation, with a focus on real-world inputs and the key concept of backpropagation. it also discusses the significance of activation functions and the structure and working of a feed forward neural network, addressing overfitting and underfitting through data division and validation.', 'chapters': [{'end': 5318.791, 'start': 5277.266, 'title': 'Understanding the f function in affine models', 'summary': "Explains the function 'f' in affine models and its role in adjusting the model based on payment data, utilizing x and y values, and emphasizing the importance of features in the model.", 'duration': 41.525, 'highlights': ["The function 'f' adjusts the model based on payment data, using x and y values for the adjustment.", "The function 'f' in affine models is called the affine thing and performs a linear operation by multiplying corresponding weights and features.", 'Emphasizing the importance of features, the value of the weight (w1) increases with the importance of the corresponding feature (x1).']}, {'end': 5712.974, 'start': 5318.791, 'title': 'Activation functions in machine learning', 'summary': "Discusses the significance of activation functions in controlling the output value 'y' in a neural network, where it emphasizes the need to bound 'y' within a specific range to avoid ambiguity and highlights the limitations of linear models in classifying data with more than three classes.", 'duration': 394.183, 'highlights': ["Activation functions are crucial in controlling the output value 'Y' in a neural network to be within a specific range, such as 0 to 1, to avoid ambiguity and ensure a particular output. The purpose of activation functions is to bound the value of 'Y' in a specific range, such as 0 to 1, to avoid ambiguity and ensure a particular output.", 'The limitations of linear models are highlighted as they are less effective in classifying data with more than three classes. Linear models are not suitable for classifying data with more than three classes as they cannot effectively divide the data into multiple classes.', "The discussion emphasizes the need to bound the output value 'Y' within a specific range to prevent ambiguity and ensure a particular output. It is crucial to bound the output value 'Y' within a specific range to avoid ambiguity and ensure a specific output."]}, {'end': 5999.774, 'start': 5713.414, 'title': 'Activation functions in machine learning', 'summary': 'Explains the importance of activation functions in introducing non-linearity into data for better classification, and controlling the output values within a specific range, essential for both classification and regression problems in machine learning.', 'duration': 286.36, 'highlights': ['Introduction of third dimension for linear separation The speaker introduced a third dimension to achieve linear separation by introducing a hyperplane, allowing for easier classification of data into separate classes.', "Purpose of activation functions Activation functions are used to introduce non-linearity in a linear model, enhancing the model's capability to classify data accurately as the number of classes increases.", 'Explanation of different activation functions The speaker provides an overview of various activation functions such as sigmoid, tanh, ReLU, and step function, and explains their specific characteristics and boundaries.', 'Control of output values within a specific range The chapter emphasizes the need to control output values within a specific range (0 to 1) for classification problems and regression problems in machine learning, ensuring accurate predictions and classifications.']}, {'end': 7003.308, 'start': 5999.774, 'title': 'Understanding feed forward neural networks', 'summary': 'Explains the structure and working of a feed forward neural network, emphasizing the importance of connecting hidden nodes to all inputs, determining the number of layers and neurons, and addressing overfitting and underfitting through data division and validation. it also highlights the significance of adjusting model capacity for optimal performance.', 'duration': 1003.534, 'highlights': ['The chapter explains the structure and working of a feed forward neural network, emphasizing the importance of connecting hidden nodes to all inputs. The feed forward neural network structure is explained, highlighting the importance of connecting hidden nodes to all inputs for effective functioning.', "Determining the number of layers and neurons is crucial for optimal performance, and the process involves hit-and-try as well as understanding the model's capacity. Determining the optimal number of layers and neurons is crucial, involving hit-and-try and understanding the model's capacity for efficient performance.", 'Addressing overfitting and underfitting is essential through data division into training, validation, and testing sets, and evaluating model performance using confusion matrix. Addressing overfitting and underfitting is essential through data division and evaluating model performance using confusion matrix.', 'Adjusting model capacity for optimal performance is crucial, and experience and prior knowledge play a significant role in developing the intuition for building effective neural networks. Adjusting model capacity for optimal performance is crucial, with experience and prior knowledge contributing to developing intuition for building effective neural networks.']}, {'end': 7621.626, 'start': 7005.939, 'title': 'Neural network basics', 'summary': 'Explains the concept of neural networks, focusing on the input layer, hidden layers, and backpropagation, with an emphasis on the number of inputs in real-world scenarios and the concept of backpropagation.', 'duration': 615.687, 'highlights': ['The number of inputs for a small image of 28x28 is 784, and for larger images like 500x500, it can reach 2.5 million, with potential multiplication for multiple images.', 'Multi-layer perceptron architecture involves multiple hidden layers, combining linear and non-linear functions, allowing for more complex calculations and modeling.', "The concept of backpropagation involves adjusting the weights of the neural network based on the error between the actual output and the predicted output, aiming to minimize the error and improve the model's accuracy."]}], 'duration': 2344.36, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/W7N6LPp0SmY/pics/W7N6LPp0SmY5277266.jpg', 'highlights': ["The concept of backpropagation involves adjusting the weights of the neural network based on the error between the actual output and the predicted output, aiming to minimize the error and improve the model's accuracy.", "Determining the optimal number of layers and neurons is crucial, involving hit-and-try and understanding the model's capacity for efficient performance.", 'Addressing overfitting and underfitting is essential through data division and evaluating model performance using confusion matrix.', "The function 'f' in affine models is called the affine thing and performs a linear operation by multiplying corresponding weights and features.", 'The feed forward neural network structure is explained, highlighting the importance of connecting hidden nodes to all inputs for effective functioning.']}, {'end': 9613.915, 'segs': [{'end': 7675.973, 'src': 'embed', 'start': 7644.498, 'weight': 1, 'content': [{'end': 7649.551, 'text': 'now the range of lab images is from minus 128 to 128..', 'start': 7644.498, 'duration': 5.053}, {'end': 7651.052, 'text': 'So here my task.', 'start': 7649.551, 'duration': 1.501}, {'end': 7655.816, 'text': 'so if generally, if you see the activation function, if you remember from the previous class,', 'start': 7651.052, 'duration': 4.764}, {'end': 7665.764, 'text': 'all our activation function from ranging something from 0 to 1 or like minus 1 to plus 1, right?', 'start': 7655.816, 'duration': 9.948}, {'end': 7675.973, 'text': 'Now I wanted something that has to go from minus 128 till plus 128..', 'start': 7666.405, 'duration': 9.568}], 'summary': 'The task is to create an activation function ranging from -128 to 128.', 'duration': 31.475, 'max_score': 7644.498, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/W7N6LPp0SmY/pics/W7N6LPp0SmY7644498.jpg'}, {'end': 7751.061, 'src': 'embed', 'start': 7723.221, 'weight': 0, 'content': [{'end': 7734.733, 'text': 'now, back propagation is one of the most used, or i should say okay, now visible, okay, now, this is one of the most not used.', 'start': 7723.221, 'duration': 11.512}, {'end': 7740.116, 'text': 'but this is one of the most obvious things in back propagation.', 'start': 7734.733, 'duration': 5.383}, {'end': 7741.957, 'text': 'this is in AI.', 'start': 7740.116, 'duration': 1.841}, {'end': 7743.297, 'text': 'and why this is famous?', 'start': 7741.957, 'duration': 1.34}, {'end': 7745.278, 'text': 'because this is the main backbone for AI.', 'start': 7743.297, 'duration': 1.981}, {'end': 7749.08, 'text': 'this is how AI learns from, learns itself, right?', 'start': 7745.278, 'duration': 3.802}, {'end': 7751.061, 'text': 'so people say that AI is learning from itself.', 'start': 7749.08, 'duration': 1.981}], 'summary': "Back propagation is a fundamental and widely used method in ai, serving as the main backbone for ai's self-learning capabilities.", 'duration': 27.84, 'max_score': 7723.221, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/W7N6LPp0SmY/pics/W7N6LPp0SmY7723221.jpg'}, {'end': 7884.15, 'src': 'embed', 'start': 7856.1, 'weight': 2, 'content': [{'end': 7866.84, 'text': 'so you can easily see here that, uh, that the error was very high at this particular point, but then it slowly, slowly, slowly decreased.', 'start': 7856.1, 'duration': 10.74}, {'end': 7869.882, 'text': "it decreased till here, but now it's almost saturated.", 'start': 7866.84, 'duration': 3.042}, {'end': 7871.763, 'text': "so it's like almost saturating return.", 'start': 7869.882, 'duration': 1.881}, {'end': 7876.226, 'text': 'at one distance it will be like parallel to the x-axis.', 'start': 7871.763, 'duration': 4.463}, {'end': 7878.287, 'text': 'it will become parallel to the x-axis.', 'start': 7876.226, 'duration': 2.061}, {'end': 7884.15, 'text': 'this is how it looks, right, shivan.', 'start': 7878.287, 'duration': 5.863}], 'summary': 'The error decreased slowly till a point, then almost saturated, resembling a parallel to the x-axis.', 'duration': 28.05, 'max_score': 7856.1, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/W7N6LPp0SmY/pics/W7N6LPp0SmY7856100.jpg'}, {'end': 7946.488, 'src': 'embed', 'start': 7914.785, 'weight': 3, 'content': [{'end': 7928.25, 'text': 'uh, all these things are like, uh, handled by the, uh, the framework itself, three times, we have to go back and forth so that the it is minimized.', 'start': 7914.785, 'duration': 13.465}, {'end': 7929.21, 'text': 'this you have to decide.', 'start': 7928.25, 'duration': 0.96}, {'end': 7935.833, 'text': "you'll see the error graph continuously and then you'll decide okay, now i want to stop or okay, no, i still i want to go further.", 'start': 7929.21, 'duration': 6.623}, {'end': 7936.933, 'text': 'this is up to your hands.', 'start': 7935.833, 'duration': 1.1}, {'end': 7944.806, 'text': 'okay, make sense, right.', 'start': 7942.784, 'duration': 2.022}, {'end': 7946.488, 'text': 'I mean, see what.', 'start': 7944.806, 'duration': 1.682}], 'summary': 'Framework handles processes, requiring three cycles to minimize errors. user decides when to stop or continue based on error graph.', 'duration': 31.703, 'max_score': 7914.785, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/W7N6LPp0SmY/pics/W7N6LPp0SmY7914785.jpg'}, {'end': 8058.75, 'src': 'embed', 'start': 8030.72, 'weight': 4, 'content': [{'end': 8037.223, 'text': 'for every corresponding else you have a weighted and these weightages are by default randomly assigned to you.', 'start': 8030.72, 'duration': 6.503}, {'end': 8043.959, 'text': "right now they're randomly assigned, assigned at the very first point, when the problem start at that particular instance.", 'start': 8037.223, 'duration': 6.736}, {'end': 8046.06, 'text': 'these are like randomly assigned to you.', 'start': 8043.959, 'duration': 2.101}, {'end': 8053.226, 'text': 'now for every input in the training data, the ANN is activated and the and its output is also.', 'start': 8046.06, 'duration': 7.166}, {'end': 8058.75, 'text': "this is what I told you, also right, that once we'll do all the calculations, we'll do the first feed forward neural network.", 'start': 8053.226, 'duration': 5.524}], 'summary': 'Weightage is randomly assigned to each corresponding else. ann is activated for each input in the training data.', 'duration': 28.03, 'max_score': 8030.72, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/W7N6LPp0SmY/pics/W7N6LPp0SmY8030720.jpg'}, {'end': 8150.942, 'src': 'embed', 'start': 8126.269, 'weight': 5, 'content': [{'end': 8131.894, 'text': 'you read the line to be there, the error is noted and the weights are adjusted accordingly.', 'start': 8126.269, 'duration': 5.625}, {'end': 8139.92, 'text': 'this process is repeated till the output error is a predetermined threshold value, or where you think that it is being saturated right.', 'start': 8131.894, 'duration': 8.026}, {'end': 8141.061, 'text': 'this is what I also told you.', 'start': 8139.92, 'duration': 1.141}, {'end': 8150.942, 'text': "now. if I tell you that how it goes, okay, let's see one by one, mathematically, I'll show you one by one also.", 'start': 8142.577, 'duration': 8.365}], 'summary': 'Weights are adjusted iteratively to minimize error until it reaches a threshold value or saturation point.', 'duration': 24.673, 'max_score': 8126.269, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/W7N6LPp0SmY/pics/W7N6LPp0SmY8126269.jpg'}, {'end': 8503.089, 'src': 'embed', 'start': 8473.62, 'weight': 6, 'content': [{'end': 8476.281, 'text': 'this is the equation of error in front of you now.', 'start': 8473.62, 'duration': 2.661}, {'end': 8477.262, 'text': 'what you can do is now.', 'start': 8476.281, 'duration': 0.981}, {'end': 8480.744, 'text': 'you might be thinking that, okay, this error equation is just a simple linear equation.', 'start': 8477.262, 'duration': 3.482}, {'end': 8482.465, 'text': 'I can solve it linearly now.', 'start': 8480.744, 'duration': 1.721}, {'end': 8490.458, 'text': 'imagine that instead of X 1, X 2, you have thousand X n right now, this equation is not a 2d equation.', 'start': 8482.465, 'duration': 7.993}, {'end': 8492.62, 'text': 'now this is a thousand dimension equation.', 'start': 8490.458, 'duration': 2.162}, {'end': 8497.244, 'text': 'now you cannot just simply do things and just check it around right.', 'start': 8492.62, 'duration': 4.624}, {'end': 8500.046, 'text': 'you cannot just do it like that you have to.', 'start': 8497.244, 'duration': 2.802}, {'end': 8503.089, 'text': 'you cannot just plot a graph and you can look into that, right.', 'start': 8500.046, 'duration': 3.043}], 'summary': 'The error equation transforms into a 1000-dimensional equation, requiring a different approach.', 'duration': 29.469, 'max_score': 8473.62, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/W7N6LPp0SmY/pics/W7N6LPp0SmY8473620.jpg'}, {'end': 8677.039, 'src': 'embed', 'start': 8649.474, 'weight': 7, 'content': [{'end': 8654.579, 'text': 'if y a is equal to equal to, then we are saying that the error should be equal to zero.', 'start': 8649.474, 'duration': 5.105}, {'end': 8659.724, 'text': 'right, error should be equal to zero, or tending towards zero, right.', 'start': 8654.579, 'duration': 5.145}, {'end': 8667.396, 'text': 'But if I say mathematically, zero is, I mean if I say statistically, this error is never going to be zero right?', 'start': 8660.134, 'duration': 7.262}, {'end': 8669.797, 'text': 'This is never even tending to zero.', 'start': 8667.576, 'duration': 2.221}, {'end': 8677.039, 'text': 'So what our main focus is, our main focus is that if E is here, our main motto is to minimize the current error.', 'start': 8669.817, 'duration': 7.222}], 'summary': 'Minimize the current error to approach zero in statistics and mathematics.', 'duration': 27.565, 'max_score': 8649.474, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/W7N6LPp0SmY/pics/W7N6LPp0SmY8649474.jpg'}, {'end': 8978.791, 'src': 'embed', 'start': 8950.25, 'weight': 8, 'content': [{'end': 8954.452, 'text': 'okay, now, with the first derivative, you can know that where is the slope of change?', 'start': 8950.25, 'duration': 4.202}, {'end': 8961.134, 'text': 'but if you want to know that which is what it is, if it is a maxima or if it is a minima, how do you know it?', 'start': 8954.452, 'duration': 6.682}, {'end': 8965.836, 'text': 'you know by doing a double differentiation of that thing and then checking its dimensions.', 'start': 8961.134, 'duration': 4.702}, {'end': 8973.887, 'text': 'if the double derivative, if the double derivative of any of your function, is greater than 0,', 'start': 8966.46, 'duration': 7.427}, {'end': 8978.791, 'text': 'if the double derivation of a function is greater than 0, then it is a minima.', 'start': 8973.887, 'duration': 4.904}], 'summary': "By taking the double derivative, if it's >0, then it's a minima.", 'duration': 28.541, 'max_score': 8950.25, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/W7N6LPp0SmY/pics/W7N6LPp0SmY8950250.jpg'}, {'end': 9065.629, 'src': 'embed', 'start': 9037.597, 'weight': 9, 'content': [{'end': 9042.059, 'text': 'And so slowly what we do is now understand this mechanism.', 'start': 9037.597, 'duration': 4.462}, {'end': 9047.482, 'text': "This is this whole thing that I'm showing you right now is called gradient descent or GD.", 'start': 9042.5, 'duration': 4.982}, {'end': 9054.146, 'text': 'right. you might have also heard this as SGD, stochastic gradient descent.', 'start': 9049.065, 'duration': 5.081}, {'end': 9058.807, 'text': 'when you do a gradient descent in batches, it is called stochastic gradient descent.', 'start': 9054.146, 'duration': 4.661}, {'end': 9060.948, 'text': "okay, I'll take a pause here.", 'start': 9058.807, 'duration': 2.141}, {'end': 9065.629, 'text': 'and to anyone who has not understood this thing, please ask.', 'start': 9060.948, 'duration': 4.681}], 'summary': 'Explaining gradient descent and stochastic gradient descent in batches.', 'duration': 28.032, 'max_score': 9037.597, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/W7N6LPp0SmY/pics/W7N6LPp0SmY9037597.jpg'}, {'end': 9314.853, 'src': 'embed', 'start': 9273.328, 'weight': 10, 'content': [{'end': 9276.43, 'text': 'now, what will be the adjusting values of w1 and w2?', 'start': 9273.328, 'duration': 3.102}, {'end': 9279.171, 'text': 'so now we already have the older values of w1, w2.', 'start': 9276.43, 'duration': 2.741}, {'end': 9298.868, 'text': 'so the new value of w1 w2 will be w is equal to w1, minus d e by d w1 and w2 will be equal to w2, minus d e by d w 2.', 'start': 9279.171, 'duration': 19.697}, {'end': 9300.929, 'text': 'right, this will be equal to this.', 'start': 9298.868, 'duration': 2.061}, {'end': 9305.53, 'text': 'and but there is a special term that is called alpha.', 'start': 9300.929, 'duration': 4.601}, {'end': 9307.491, 'text': 'that is associated here.', 'start': 9305.53, 'duration': 1.961}, {'end': 9309.691, 'text': 'it is multiplied here.', 'start': 9307.491, 'duration': 2.2}, {'end': 9314.853, 'text': "i'll tell you what alpha is here.", 'start': 9309.691, 'duration': 5.162}], 'summary': 'Updating w1 and w2 using alpha, de/dw1, and de/dw2', 'duration': 41.525, 'max_score': 9273.328, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/W7N6LPp0SmY/pics/W7N6LPp0SmY9273328.jpg'}, {'end': 9463.641, 'src': 'embed', 'start': 9433.934, 'weight': 11, 'content': [{'end': 9439.218, 'text': 'also right, your error is also going to shift very, very slowly because you have made a very small change in w1.', 'start': 9433.934, 'duration': 5.284}, {'end': 9443.947, 'text': "otherwise, let's assume that you might there.", 'start': 9440.484, 'duration': 3.463}, {'end': 9449.251, 'text': 'your w1 originally was here and you made a very big change in W.', 'start': 9443.947, 'duration': 5.304}, {'end': 9457.317, 'text': 'so your error might be here at this point, new error might be here, or it might also happen that it might come here.', 'start': 9449.251, 'duration': 8.066}, {'end': 9463.641, 'text': "also right, it might change that how gradually you're moving your W.", 'start': 9457.317, 'duration': 6.324}], 'summary': 'Small change in w1 leads to slow error shift, while big change in w can cause error to shift suddenly.', 'duration': 29.707, 'max_score': 9433.934, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/W7N6LPp0SmY/pics/W7N6LPp0SmY9433934.jpg'}, {'end': 9585.63, 'src': 'embed', 'start': 9553.01, 'weight': 12, 'content': [{'end': 9561.974, 'text': 'so you have to take a very uh, critical value of alpha such that it moves, uh, uh, very uh, slowly.', 'start': 9553.01, 'duration': 8.964}, {'end': 9563.836, 'text': 'i mean not slowly.', 'start': 9561.974, 'duration': 1.862}, {'end': 9568.059, 'text': 'yeah, are you trying to like tune up your model?', 'start': 9563.836, 'duration': 4.223}, {'end': 9570.401, 'text': 'uh, this is not tuning, it is just adjusting.', 'start': 9568.059, 'duration': 2.342}, {'end': 9576.285, 'text': 'the the thing that how the model is reacting to the changes that we are that we are doing.', 'start': 9570.401, 'duration': 5.884}, {'end': 9585.63, 'text': "so it's like we are controlling our own process of the flow of gradient and alpha.", 'start': 9576.285, 'duration': 9.345}], 'summary': "Adjust alpha to control model's gradient flow and reactions to changes.", 'duration': 32.62, 'max_score': 9553.01, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/W7N6LPp0SmY/pics/W7N6LPp0SmY9553010.jpg'}], 'start': 7621.626, 'title': 'Neural network training and optimization', 'summary': 'Covers neural network activation, backpropagation, training process, gradient descent, error minimization, and the adjustment of weights, emphasizing the role of human decision-making, mathematical concepts, and the importance of controlling the learning rate for optimization.', 'chapters': [{'end': 8011.589, 'start': 7621.626, 'title': 'Neural network activation and backpropagation', 'summary': 'Covers the concept of lab images in neural network activation functions, the iterative process of backpropagation, and the manual vs automated handling of the process, emphasizing the importance of human decision-making in stopping or continuing the error minimization process.', 'duration': 389.963, 'highlights': ['Backpropagation is one of the most important processes in AI and serves as the main backbone for AI, as it allows the system to learn from its errors. Backpropagation is crucial for AI and serves as the main backbone for learning from errors.', 'The lab images have a range from -128 to 128, which is different from the typical activation function range of 0 to 1 or -1 to 1, requiring a customized activation function to accommodate the wider range. Lab images have a range from -128 to 128, necessitating a custom activation function to handle the wider range.', 'The iterative process of backpropagation involves adjusting the values of w1 and w2 multiple times to minimize the error, eventually leading to the error graph saturating and becoming parallel to the x-axis. Backpropagation involves iterative adjustment of w1 and w2 values to minimize error, leading to the error graph saturating and becoming parallel to the x-axis.', 'While the framework handles the back and forth process in error minimization, the decision of when to stop or continue the process remains in human hands, emphasizing the importance of human decision-making in the process. The framework handles the back and forth process in error minimization, but the decision of when to stop or continue the process remains in human hands.']}, {'end': 8532.242, 'start': 8012.512, 'title': 'Neural network training process', 'summary': 'Explains the process of training a neural network, including the concept of weighted edges, activation of the ann, comparison of output with ground truth, error propagation, and minimization of the error equation in a multi-dimensional space.', 'duration': 519.73, 'highlights': ['The process of training a neural network involves weighted edges where each edge has a corresponding weighted rate, and these weightages are randomly assigned at the start of the problem.', 'The ANN is activated for every input in the training data, and its output is compared with the desired output, known as ground truth, to minimize the error equation and achieve a predetermined threshold value or saturation point.', 'Error propagation occurs by noting the error and adjusting the weights accordingly, and the process is repeated until the output error reaches a predetermined threshold value or saturation point.', 'The error equation, initially a simple linear equation, becomes a complex multi-dimensional equation in the context of neural network training, making it challenging to minimize the error and necessitating a more complex approach than linear solving.']}, {'end': 9142.677, 'start': 8532.242, 'title': 'Gradient descent in deep learning', 'summary': 'Discusses the process of minimizing the error in a deep learning model using gradient descent technique, focusing on the mathematical concepts of first and double differentiation to find the global minima and the role of gradient descent in deep learning.', 'duration': 610.435, 'highlights': ['The error in the deep learning model is minimized by the process of gradient descent, aiming to achieve a global minima by reducing the error to zero or tending towards zero. Minimizing the error in the model, aiming to reduce it to zero or tending towards zero.', 'The mathematical process of first and double differentiation is used to determine the points of change in slope and identify whether it is a minima or maxima. Using first and double differentiation to identify the points of change in slope and determine if it is a minima or maxima.', 'The process of gradient descent, also known as stochastic gradient descent in batches, is instrumental in finding the global minima in deep learning models. The role of gradient descent, particularly stochastic gradient descent, in finding the global minima in deep learning.']}, {'end': 9373.586, 'start': 9142.677, 'title': 'Derivative of e and adjustment of w1 and w2', 'summary': 'Discusses the process of taking the derivative of e with respect to w1 and w2, and the adjustment of new values for w1 and w2 using the learning rate alpha.', 'duration': 230.909, 'highlights': ['The new values of w1 and w2 will be w is equal to w1 minus d e by d w1 and w2 will be equal to w2 minus d e by d w 2, multiplied by the learning rate alpha.', 'Taking the derivative of E with respect to W1 results in X1, and with respect to W2 results in X2, both multiplied by a constant.', 'The chapter emphasizes that the constant term is equal to zero and does not bring much value.', 'The learning rate alpha is associated with the adjustment of the new values for w1 and w2, and is multiplied in the process.']}, {'end': 9613.915, 'start': 9373.586, 'title': 'Gradient descent and learning rate', 'summary': 'Discusses the importance of controlling the learning rate (alpha) in gradient descent to avoid overshooting the global minima and emphasizes the need for a critical value of alpha to balance between convergence speed and accuracy.', 'duration': 240.329, 'highlights': ['The error will shift very slowly with a small change in w1, emphasizing the importance of moving the value of w slowly. Emphasizes the impact of making small changes in w1 on the shifting of the error.', 'The need to control the learning rate (alpha) to avoid missing the global minima, highlighting the significance of selecting a suitable alpha value. Emphasizes the importance of controlling the learning rate to prevent overshooting the global minima.', 'The discussion on selecting a critical value of alpha to balance between convergence speed and accuracy in gradient descent. Emphasizes the need for a critical value of alpha to balance between convergence speed and accuracy.']}], 'duration': 1992.289, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/W7N6LPp0SmY/pics/W7N6LPp0SmY7621626.jpg', 'highlights': ['Backpropagation is crucial for AI and serves as the main backbone for learning from errors.', 'Lab images have a range from -128 to 128, necessitating a custom activation function to handle the wider range.', 'The iterative adjustment of w1 and w2 values minimizes error, leading to the error graph saturating and becoming parallel to the x-axis.', 'The framework handles the back and forth process in error minimization, but the decision of when to stop or continue the process remains in human hands.', 'The process of training a neural network involves weighted edges with randomly assigned weightages at the start.', 'Error propagation occurs by noting the error and adjusting the weights until the output error reaches a predetermined threshold value or saturation point.', 'The error equation becomes a complex multi-dimensional equation in the context of neural network training, necessitating a more complex approach than linear solving.', 'Minimizing the error in the model, aiming to reduce it to zero or tending towards zero.', 'Using first and double differentiation to identify the points of change in slope and determine if it is a minima or maxima.', 'The role of gradient descent, particularly stochastic gradient descent, in finding the global minima in deep learning.', 'The new values of w1 and w2 are adjusted by the learning rate alpha in the process.', 'Emphasizes the impact of making small changes in w1 on the shifting of the error.', 'Emphasizes the importance of controlling the learning rate to prevent overshooting the global minima.', 'Emphasizes the need for a critical value of alpha to balance between convergence speed and accuracy.']}, {'end': 11276.092, 'segs': [{'end': 9649.599, 'src': 'embed', 'start': 9614.275, 'weight': 0, 'content': [{'end': 9618.561, 'text': 'It depends on your data that how slowly you see that your model is moving.', 'start': 9614.275, 'duration': 4.286}, {'end': 9621.044, 'text': 'This is just the initial value that you should take.', 'start': 9618.981, 'duration': 2.063}, {'end': 9631.701, 'text': 'So if you will see all of your major frameworks like Keras or like TensorFlow, you will see that they have put a value of 0.001 also by default.', 'start': 9622.066, 'duration': 9.635}, {'end': 9632.162, 'text': 'Make sense?', 'start': 9631.741, 'duration': 0.421}, {'end': 9649.599, 'text': 'till here.', 'start': 9648.919, 'duration': 0.68}], 'summary': 'Model training speed depends on data and initial learning rate, often set at 0.001.', 'duration': 35.324, 'max_score': 9614.275, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/W7N6LPp0SmY/pics/W7N6LPp0SmY9614275.jpg'}, {'end': 9885.527, 'src': 'embed', 'start': 9781.887, 'weight': 1, 'content': [{'end': 9791.812, 'text': 'so i told you that this image is nothing, but images are just numbers, numbers ranging, numbers ranging from zero till two, five, five,', 'start': 9781.887, 'duration': 9.925}, {'end': 9797.596, 'text': 'where 255 being the highest value and 0 being the lowest value.', 'start': 9794.093, 'duration': 3.503}, {'end': 9800.558, 'text': 'And these are called pixel values.', 'start': 9798.036, 'duration': 2.522}, {'end': 9804.621, 'text': 'These are pixel values, VI values.', 'start': 9802.079, 'duration': 2.542}, {'end': 9809.125, 'text': 'These are pixel values to you.', 'start': 9805.542, 'duration': 3.583}, {'end': 9810.066, 'text': 'These are pixel values.', 'start': 9809.185, 'duration': 0.881}, {'end': 9811.827, 'text': 'This is how the images work.', 'start': 9810.106, 'duration': 1.721}, {'end': 9816.15, 'text': 'But the image that you are seeing on your screen, this is a single-channel image.', 'start': 9811.887, 'duration': 4.263}, {'end': 9817.591, 'text': 'This is a single-channel image.', 'start': 9816.17, 'duration': 1.421}, {'end': 9821.014, 'text': 'Now, you might have multi-channel images also.', 'start': 9818.012, 'duration': 3.002}, {'end': 9822.735, 'text': 'Let me show you the demo.', 'start': 9821.034, 'duration': 1.701}, {'end': 9836.687, 'text': 'okay, well, last time anyone in doubts in back propagation, because this ends your theory for uh, uh, your theory for deep, uh for neural networks.', 'start': 9824.184, 'duration': 12.503}, {'end': 9849.871, 'text': "right, neural network intro is done here, right, i'll take up a new python and i'll just demonstrate you how.", 'start': 9836.687, 'duration': 13.184}, {'end': 9877.94, 'text': 'images. okay, there are multiple libraries that you might want to use these libraries for image reading, image processing, and i personally feel, if you, uh,', 'start': 9849.871, 'duration': 28.069}, {'end': 9885.527, 'text': 'if you, if you, if you have passion for images, these are the best libraries that you should use And see.', 'start': 9877.94, 'duration': 7.587}], 'summary': 'Images are represented by pixel values ranging from 0 to 255, and can be single or multi-channel. libraries are available for image processing.', 'duration': 103.64, 'max_score': 9781.887, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/W7N6LPp0SmY/pics/W7N6LPp0SmY9781887.jpg'}, {'end': 10510.27, 'src': 'embed', 'start': 10447.011, 'weight': 5, 'content': [{'end': 10450.552, 'text': 'we obviously know this is a dimension, but now we have a third dimension.', 'start': 10447.011, 'duration': 3.541}, {'end': 10452.473, 'text': 'this is called number of channels.', 'start': 10450.552, 'duration': 1.921}, {'end': 10455.254, 'text': 'so how many number of channels does your image has?', 'start': 10452.473, 'duration': 2.781}, {'end': 10458.535, 'text': 'so this image has three channel images.', 'start': 10455.254, 'duration': 3.281}, {'end': 10460.375, 'text': 'so rgb are separated.', 'start': 10458.535, 'duration': 1.84}, {'end': 10466.077, 'text': "but let's say i took this example where i read the image with a zero tag.", 'start': 10460.375, 'duration': 5.702}, {'end': 10470.803, 'text': "now zero tag means that you don't, you just want this image to be in a single channel image.", 'start': 10466.721, 'duration': 4.082}, {'end': 10471.443, 'text': 'so what we did?', 'start': 10470.803, 'duration': 0.64}, {'end': 10472.724, 'text': 'it converted this image.', 'start': 10471.443, 'duration': 1.281}, {'end': 10491.433, 'text': 'see only single channels, right, make sense, okay.', 'start': 10472.724, 'duration': 18.709}, {'end': 10495.226, 'text': 'so now sometimes people confuse that.', 'start': 10492.845, 'duration': 2.381}, {'end': 10497.446, 'text': 'you know great and gray.', 'start': 10495.226, 'duration': 2.22}, {'end': 10503.748, 'text': 'oh, this is a single channel image and people think that single channel images is a black and white image.', 'start': 10497.446, 'duration': 6.302}, {'end': 10507.83, 'text': 'no black and white images, when you will only have black and white either 0 or 255.', 'start': 10503.748, 'duration': 4.082}, {'end': 10510.27, 'text': 'this is a grayscale image.', 'start': 10507.83, 'duration': 2.44}], 'summary': 'Image has 3 channels (rgb). single channel can be grayscale (0-255).', 'duration': 63.259, 'max_score': 10447.011, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/W7N6LPp0SmY/pics/W7N6LPp0SmY10447011.jpg'}, {'end': 10677.983, 'src': 'embed', 'start': 10652.957, 'weight': 3, 'content': [{'end': 10661.161, 'text': "and especially with the libraries like opencv, which are now using deep learning to do image processing, it's like it's like a whole circle.", 'start': 10652.957, 'duration': 8.204}, {'end': 10665.564, 'text': 'so you do, you do image processing because you want to do deep learn.', 'start': 10661.161, 'duration': 4.403}, {'end': 10667.044, 'text': 'you want to do deep learning, right.', 'start': 10665.564, 'duration': 1.48}, {'end': 10677.983, 'text': 'so this is like a circle like this you want to do image processing because ultimately you want to do deep learning, what people are doing.', 'start': 10667.044, 'duration': 10.939}], 'summary': 'Libraries like opencv are using deep learning for image processing, creating a circular relationship between the two technologies.', 'duration': 25.026, 'max_score': 10652.957, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/W7N6LPp0SmY/pics/W7N6LPp0SmY10652957.jpg'}, {'end': 10751.76, 'src': 'embed', 'start': 10707.314, 'weight': 8, 'content': [{'end': 10713.996, 'text': "you just say pip, install opencv, hyphen, Python, it's not cv2 right.", 'start': 10707.314, 'duration': 6.682}, {'end': 10722.038, 'text': 'and then, when you, then you say Python and then you will import it like import cv2.', 'start': 10713.996, 'duration': 8.042}, {'end': 10724.258, 'text': "so there's a little confusion.", 'start': 10722.038, 'duration': 2.22}, {'end': 10731.28, 'text': 'so you you install it by opencv, Python, but you import it as cv2 right.', 'start': 10724.258, 'duration': 7.022}, {'end': 10739.671, 'text': 'uh, similarly, for sk image, sk image is simple.', 'start': 10737.089, 'duration': 2.582}, {'end': 10742.113, 'text': 'pip, install sk image for pil.', 'start': 10739.671, 'duration': 2.442}, {'end': 10751.76, 'text': 'pil is again a problem.', 'start': 10742.113, 'duration': 9.647}], 'summary': "To install opencv, use 'pip install opencv-python', and to import it, use 'import cv2'. similarly, for scikit-image, use 'pip install scikit-image'.", 'duration': 44.446, 'max_score': 10707.314, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/W7N6LPp0SmY/pics/W7N6LPp0SmY10707314.jpg'}, {'end': 11157.626, 'src': 'embed', 'start': 11117.815, 'weight': 9, 'content': [{'end': 11125.285, 'text': 'okay, because, see, you will not see any problem right now, but you know when.', 'start': 11117.815, 'duration': 7.47}, {'end': 11132.009, 'text': 'see, i have faced this problem a lot whenever i do this with the, you know.', 'start': 11125.285, 'duration': 6.724}, {'end': 11139.333, 'text': 'uh, whenever i use a new you know framework, the problem is that you know images are read very wrongly.', 'start': 11132.009, 'duration': 7.324}, {'end': 11142.135, 'text': 'so the so the wrong images are now fed to the model.', 'start': 11139.333, 'duration': 2.802}, {'end': 11145.577, 'text': 'now this is how things get worse there.', 'start': 11142.135, 'duration': 3.442}, {'end': 11157.626, 'text': 'okay, you can see my screen right.', 'start': 11150.279, 'duration': 7.347}], 'summary': 'New framework causes image misreading, leading to model input errors.', 'duration': 39.811, 'max_score': 11117.815, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/W7N6LPp0SmY/pics/W7N6LPp0SmY11117815.jpg'}], 'start': 9614.275, 'title': 'Image processing in deep learning', 'summary': 'Introduces the initial learning rate for models, pixel values, different image channels, and essential libraries for image processing in deep learning. it discusses extracting and visualizing rgb channels, optics, image dimensions, and various image processing techniques using opencv and deep learning. emphasizes the significance of image processing and its connection to deep learning.', 'chapters': [{'end': 10079.343, 'start': 9614.275, 'title': 'Introduction to image processing in deep learning', 'summary': 'Introduces the initial learning rate for models, the pixel values of images, the different types of image channels, and the essential libraries for image processing in deep learning, emphasizing the significance of image processing in deep learning and the key libraries for image manipulation.', 'duration': 465.068, 'highlights': ["The chapter introduces the initial learning rate for models, emphasizing that major frameworks like Keras and TensorFlow default to a value of 0.001. It's highlighted that major frameworks like Keras and TensorFlow default to a learning rate value of 0.001, underlining its significance in model training.", 'The chapter explains the concept of image pixel values, ranging from 0 to 255, and highlights the significance of understanding pixel values in image processing. The concept of image pixel values, ranging from 0 to 255, is explained, emphasizing the importance of understanding these values in image processing.', 'The chapter discusses the different types of image channels, distinguishing between single-channel and multi-channel images, and emphasizes the significance of understanding image channels in image processing. The distinction between single-channel and multi-channel images is discussed, highlighting the importance of understanding image channels in image processing.', 'The chapter emphasizes the importance of libraries such as OpenCV, scikit-image, and PIL for image reading and processing, particularly in the context of deep learning and image manipulation. The significance of libraries like OpenCV, scikit-image, and PIL for image reading and processing is emphasized, particularly in the context of deep learning and image manipulation.', 'The chapter demonstrates the process of importing and visualizing images using OpenCV and matplotlib libraries, providing insight into the practical aspects of image processing in deep learning. The process of importing and visualizing images using OpenCV and matplotlib libraries is demonstrated, providing practical insight into image processing in deep learning.']}, {'end': 10380.632, 'start': 10079.343, 'title': 'Understanding image channels in python', 'summary': 'Discusses the process of extracting and visualizing individual rgb channels from an image, illustrating how to convert a three-channel image into separate r, g, and b channels and displaying them using python libraries like opencv and matplotlib.', 'duration': 301.289, 'highlights': ['The process of extracting individual RGB channels from an image and visualizing them using Python libraries like OpenCV and Matplotlib is discussed, highlighting the conversion of a three-channel image into separate R, G, and B channels.', 'The method of converting a three-channel image into separate R, G, and B channels and displaying them using Python libraries like OpenCV and Matplotlib is explained, emphasizing the visualization of individual RGB channels.', 'The explanation of how to convert a three-channel image into separate R, G, and B channels and visualize them using Python libraries like OpenCV and Matplotlib is provided, demonstrating the process of extracting and displaying individual RGB channels.']}, {'end': 10626.567, 'start': 10380.632, 'title': 'Understanding optics and image dimensions', 'summary': 'Explains the reflection of primary colors based on wavelength properties, the creation of final images using rgb channels, and the distinction between single channel and grayscale images in image processing.', 'duration': 245.935, 'highlights': ['The chapter explains the reflection of primary colors based on wavelength properties. It describes how primary colors (RGB) reflect back based on their wavelength properties to create the final image.', 'The creation of final images using RGB channels is discussed. It explains how the final image is created by adding the values of red, green, and blue channels together.', 'The distinction between single channel and grayscale images in image processing is clarified. It distinguishes between single channel images and grayscale images, emphasizing that grayscale images have pixel values ranging from 0 to 255 in a single channel without separate RGB values.']}, {'end': 11276.092, 'start': 10626.567, 'title': 'Image processing techniques with opencv and deep learning', 'summary': 'Discusses various image processing techniques using opencv and deep learning, emphasizing the circular relationship between image processing and deep learning, the installation process for opencv, and the importance of properly reading and handling images to avoid feeding incorrect data to models.', 'duration': 649.525, 'highlights': ['The circular relationship between image processing and deep learning is emphasized, with a mention of using deep learning for image processing and vice versa. Emphasizes the circular relationship between image processing and deep learning, mentioning the use of deep learning for image processing and vice versa.', "The installation process for OpenCV is explained, including the command 'pip install opencv-python' and the necessity to import it as 'cv2'. Explains the installation process for OpenCV, including the command 'pip install opencv-python' and the necessity to import it as 'cv2'.", 'The importance of properly reading and handling images to avoid feeding incorrect data to models is discussed, highlighting the potential issues of reading images wrongly when using new frameworks. Discusses the importance of properly reading and handling images to avoid feeding incorrect data to models, highlighting the potential issues of reading images wrongly when using new frameworks.']}], 'duration': 1661.817, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/W7N6LPp0SmY/pics/W7N6LPp0SmY9614275.jpg', 'highlights': ['Major frameworks like Keras and TensorFlow default to a learning rate value of 0.001, underlining its significance in model training.', 'The concept of image pixel values, ranging from 0 to 255, is explained, emphasizing the importance of understanding these values in image processing.', 'The distinction between single-channel and multi-channel images is discussed, highlighting the importance of understanding image channels in image processing.', 'The significance of libraries like OpenCV, scikit-image, and PIL for image reading and processing is emphasized, particularly in the context of deep learning and image manipulation.', 'The process of importing and visualizing images using OpenCV and matplotlib libraries is demonstrated, providing practical insight into image processing in deep learning.', 'The method of converting a three-channel image into separate R, G, and B channels and displaying them using Python libraries like OpenCV and Matplotlib is explained, emphasizing the visualization of individual RGB channels.', 'It distinguishes between single channel images and grayscale images, emphasizing that grayscale images have pixel values ranging from 0 to 255 in a single channel without separate RGB values.', 'Emphasizes the circular relationship between image processing and deep learning, mentioning the use of deep learning for image processing and vice versa.', "Explains the installation process for OpenCV, including the command 'pip install opencv-python' and the necessity to import it as 'cv2'.", 'Discusses the importance of properly reading and handling images to avoid feeding incorrect data to models, highlighting the potential issues of reading images wrongly when using new frameworks.']}, {'end': 12439.681, 'segs': [{'end': 11341.253, 'src': 'embed', 'start': 11311.738, 'weight': 1, 'content': [{'end': 11320.867, 'text': 'so okay, first of all, a brief introduction that what this okay, what does MNIST problem is and how does this came into power?', 'start': 11311.738, 'duration': 9.129}, {'end': 11330.787, 'text': 'okay, If we go back and if I just show you MNIST, if I just type in MNIST was actually that is National Institute for Science and Technology.', 'start': 11320.867, 'duration': 9.92}, {'end': 11331.888, 'text': 'They got up.', 'start': 11330.887, 'duration': 1.001}, {'end': 11335.45, 'text': 'So this is a problem of like early 90s.', 'start': 11333.169, 'duration': 2.281}, {'end': 11341.253, 'text': 'you know the postal department of the u.s.', 'start': 11338.132, 'duration': 3.121}], 'summary': 'Introduction to the mnist problem from the early 90s.', 'duration': 29.515, 'max_score': 11311.738, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/W7N6LPp0SmY/pics/W7N6LPp0SmY11311738.jpg'}, {'end': 11561.522, 'src': 'embed', 'start': 11536.111, 'weight': 3, 'content': [{'end': 11541.535, 'text': 'download this example from their uh, download this example from their website.', 'start': 11536.111, 'duration': 5.424}, {'end': 11544.077, 'text': 'now it will go and it will download this section.', 'start': 11541.535, 'duration': 2.542}, {'end': 11548.981, 'text': 'as of now, you can see in the panel on the left panel, there is no such thing called eminist data.', 'start': 11544.077, 'duration': 4.904}, {'end': 11557.318, 'text': "Now what I've done is I have asked it to create a folder called MNIST data right? And it will create and it will download it inside.", 'start': 11549.73, 'duration': 7.588}, {'end': 11561.522, 'text': 'I have done one more thing there is that is called one hot encoding equal to true.', 'start': 11557.778, 'duration': 3.744}], 'summary': 'Download example from website, create mnist data folder and enable one hot encoding.', 'duration': 25.411, 'max_score': 11536.111, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/W7N6LPp0SmY/pics/W7N6LPp0SmY11536111.jpg'}, {'end': 11862.197, 'src': 'embed', 'start': 11792.811, 'weight': 0, 'content': [{'end': 11810.135, 'text': 'so you I mean you already know, first is training crane, then is validation and then is testing, right.', 'start': 11792.811, 'duration': 17.324}, {'end': 11813.137, 'text': 'so 55, 000 in training, 5, 000 in validation and 10, 000.', 'start': 11810.135, 'duration': 3.002}, {'end': 11814.538, 'text': 'testing we have.', 'start': 11813.137, 'duration': 1.401}, {'end': 11818.96, 'text': 'okay, and but why is it 784?', 'start': 11814.538, 'duration': 4.422}, {'end': 11822.623, 'text': 'okay, this I understand 55, 000, because these are the number of images that we have.', 'start': 11818.96, 'duration': 3.663}, {'end': 11824.524, 'text': 'but what is this 784?', 'start': 11822.623, 'duration': 1.901}, {'end': 11831.226, 'text': 'the 784 makes sense is that your image is actually 28 cross 28.', 'start': 11824.524, 'duration': 6.702}, {'end': 11842.231, 'text': 'your image is actually 28 in height and 28 in length.', 'start': 11831.226, 'duration': 11.005}, {'end': 11844.952, 'text': 'right, so 28 cross 28.', 'start': 11842.231, 'duration': 2.721}, {'end': 11848.794, 'text': 'so now they have not given you a image into a two-dimensional array.', 'start': 11844.952, 'duration': 3.842}, {'end': 11856.938, 'text': 'what they have done is now just just listen to me, just for a second, right, this is called the process of flattening.', 'start': 11848.794, 'duration': 8.144}, {'end': 11858.799, 'text': 'this is called the process of flattening.', 'start': 11856.938, 'duration': 1.861}, {'end': 11861.016, 'text': "now let's imagine that you are here.", 'start': 11859.215, 'duration': 1.801}, {'end': 11862.197, 'text': 'you have 28 rows.', 'start': 11861.016, 'duration': 1.181}], 'summary': 'Training 55,000 images, 5,000 in validation, 784 in testing, each image is 28x28', 'duration': 69.386, 'max_score': 11792.811, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/W7N6LPp0SmY/pics/W7N6LPp0SmY11792811.jpg'}, {'end': 12308.345, 'src': 'embed', 'start': 12277.362, 'weight': 5, 'content': [{'end': 12281.123, 'text': 'But now you are asking the question that if you have multi-class classification?', 'start': 12277.362, 'duration': 3.761}, {'end': 12283.036, 'text': 'Yes, exactly.', 'start': 12281.776, 'duration': 1.26}, {'end': 12285.277, 'text': 'Multi-class classification.', 'start': 12283.797, 'duration': 1.48}, {'end': 12297.982, 'text': "And what problem do you face there? I understood what I'm trying to get at, but I did not know how to write the code.", 'start': 12288.538, 'duration': 9.444}, {'end': 12302.623, 'text': "We're trying to predict over here the one-hot code or multi-class.", 'start': 12298.942, 'duration': 3.681}, {'end': 12308.345, 'text': 'So are you going to write a multi-class predictor over here? Yes, we will write.', 'start': 12303.023, 'duration': 5.322}], 'summary': 'Discussion on writing a multi-class predictor for classification', 'duration': 30.983, 'max_score': 12277.362, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/W7N6LPp0SmY/pics/W7N6LPp0SmY12277362.jpg'}, {'end': 12400.424, 'src': 'embed', 'start': 12373.783, 'weight': 2, 'content': [{'end': 12383.187, 'text': 'that now you have to predict out of every uh category at which it belongs to which category and mns is again a multi-class classification problem.', 'start': 12373.783, 'duration': 9.404}, {'end': 12391.057, 'text': 'right where you have been given you know, 10 categories zero, one, two, three,', 'start': 12383.187, 'duration': 7.87}, {'end': 12397.041, 'text': 'four until nine and for which you have to tell that it belongs to which category.', 'start': 12391.057, 'duration': 5.984}, {'end': 12398.943, 'text': 'now, a very good question.', 'start': 12397.041, 'duration': 1.902}, {'end': 12400.424, 'text': 'now i want to answer this question.', 'start': 12398.943, 'duration': 1.481}], 'summary': 'Predict the category for a multi-class classification problem with 10 categories.', 'duration': 26.641, 'max_score': 12373.783, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/W7N6LPp0SmY/pics/W7N6LPp0SmY12373783.jpg'}], 'start': 11276.092, 'title': 'Understanding mnist and multi-class image classification', 'summary': 'Covers the history and problem statement of mnist, with a dataset of 60,000 training images and 10,000 test images, and explains one hot encoding, flattening images, and dimensions of the mnist dataset with 55,000 training images, 5,000 validation images, and 10,000 testing images, each comprising 784 pixels. it also discusses the challenges and framework used for multi-class image classification.', 'chapters': [{'end': 11557.318, 'start': 11276.092, 'title': 'Introduction to mnist problem', 'summary': 'Covers the history and problem statement of mnist, which originated as a solution to reading postal codes, leading to the creation of a dataset with 60,000 training images and 10,000 test images for machine learning validation.', 'duration': 281.226, 'highlights': ['The MNIST problem originated from the need to read postal codes efficiently, leading to the creation of a dataset with 60,000 training images and 10,000 test images. This dataset was created to address the inefficiency in reading postal codes, with 60,000 training images and 10,000 test images to facilitate machine learning validation.', 'The MNIST dataset was designed to accommodate the variability in handwritten characters, which posed a challenge for traditional character recognition methods. The variability in handwritten characters presented a challenge for traditional character recognition methods, leading to the creation of the MNIST dataset to address this issue.', 'TensorFlow provides a built-in functionality to download the MNIST dataset, simplifying the process for users. TensorFlow offers a built-in functionality to download the MNIST dataset, streamlining the process for users and enhancing accessibility.']}, {'end': 12142.333, 'start': 11557.778, 'title': 'Understanding one hot encoding in machine learning', 'summary': 'Explains the concept of one hot encoding, the process of flattening images, and the dimensions of the mnist dataset, with 55,000 training images, 5,000 validation images, and 10,000 testing images, each comprising 784 pixels.', 'duration': 584.555, 'highlights': ['The MNIST dataset consists of 55,000 training images, 5,000 validation images, and 10,000 testing images, with each image containing 784 pixels. The dataset includes 55,000 training images, 5,000 validation images, and 10,000 testing images, each comprising 784 pixels.', 'Explanation of the process of flattening images, converting 28x28 images into a 784-pixel row-major format for processing in machine learning models. The process of flattening converts 28x28 images into a 784-pixel row-major format for processing in machine learning models.', 'Insight into the concept of one hot encoding, transforming numerical labels into a binary format for machine learning model understanding and implementation. The concept of one hot encoding involves transforming numerical labels into a binary format for machine learning model understanding and implementation.']}, {'end': 12439.681, 'start': 12142.333, 'title': 'Understanding multi-class image classification', 'summary': 'Discusses the challenges faced in multi-class image classification and the distinction between binary and multi-class classification, highlighting the need for one-hot encoding and the framework used for training.', 'duration': 297.348, 'highlights': ['The distinction between binary and multi-class classification is explained, emphasizing the need for one-hot encoding and the framework used for training. The chapter delves into the difference between binary and multi-class classification, emphasizing the need for one-hot encoding and mentions the framework used for training.', 'The challenges in multi-class image classification are discussed, focusing on the difficulty of predicting which category an image belongs to. The challenges in multi-class image classification are discussed, specifically the difficulty of predicting the category to which an image belongs.', 'The need for one-hot encoding is highlighted in the context of multi-class classification, explaining the process of assigning numerical values to categories. The chapter emphasizes the need for one-hot encoding in multi-class classification, explaining the process of assigning numerical values to categories.']}], 'duration': 1163.589, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/W7N6LPp0SmY/pics/W7N6LPp0SmY11276092.jpg', 'highlights': ['The MNIST dataset consists of 55,000 training images, 5,000 validation images, and 10,000 testing images, with each image containing 784 pixels.', 'The MNIST problem originated from the need to read postal codes efficiently, leading to the creation of a dataset with 60,000 training images and 10,000 test images.', 'The challenges in multi-class image classification are discussed, focusing on the difficulty of predicting which category an image belongs to.', 'TensorFlow provides a built-in functionality to download the MNIST dataset, simplifying the process for users.', 'Explanation of the process of flattening images, converting 28x28 images into a 784-pixel row-major format for processing in machine learning models.', 'The distinction between binary and multi-class classification is explained, emphasizing the need for one-hot encoding and the framework used for training.']}, {'end': 14834.99, 'segs': [{'end': 12560.127, 'src': 'embed', 'start': 12527.521, 'weight': 0, 'content': [{'end': 12530.262, 'text': 'so, basically, you will always have a softmax.', 'start': 12527.521, 'duration': 2.741}, {'end': 12540.74, 'text': 'you will always have a softmax function, this called softmax.', 'start': 12534.177, 'duration': 6.563}, {'end': 12543.04, 'text': 'you will have a softmax activation function.', 'start': 12540.74, 'duration': 2.3}, {'end': 12545.481, 'text': 'now, what does a softmax activation function does?', 'start': 12543.04, 'duration': 2.441}, {'end': 12554.545, 'text': 'is softmax activation function, whenever you give it values, what it will make sure that whenever the two probabilities are coming out there,', 'start': 12545.481, 'duration': 9.064}, {'end': 12560.127, 'text': 'some will be always equal to one, irrespective of how many classes are here.', 'start': 12554.545, 'duration': 5.582}], 'summary': 'Softmax activation ensures sum of probabilities equals one for any number of classes.', 'duration': 32.606, 'max_score': 12527.521, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/W7N6LPp0SmY/pics/W7N6LPp0SmY12527521.jpg'}, {'end': 13009.228, 'src': 'embed', 'start': 12986.125, 'weight': 1, 'content': [{'end': 12993.768, 'text': 'before going to deep learning, before we do all this magic, i just want to show you how do you do, uh, all this stuff by your hands?', 'start': 12986.125, 'duration': 7.643}, {'end': 12995.089, 'text': 'so see all this.', 'start': 12993.768, 'duration': 1.321}, {'end': 12996.409, 'text': "you don't have to do anything, right?", 'start': 12995.089, 'duration': 1.32}, {'end': 12998.09, 'text': 'back propagation, feed forward.', 'start': 12996.409, 'duration': 1.681}, {'end': 13004.867, 'text': 'everything is happening on its own 55 thousand images are going backward forward multiple times, and you are doing nothing.', 'start': 12998.09, 'duration': 6.777}, {'end': 13006.628, 'text': 'right, you are doing nothing.', 'start': 13004.867, 'duration': 1.761}, {'end': 13009.228, 'text': "so so you're just sitting and you just gave it.", 'start': 13006.628, 'duration': 2.6}], 'summary': 'Deep learning automates processes with thousands of images, requiring minimal human intervention.', 'duration': 23.103, 'max_score': 12986.125, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/W7N6LPp0SmY/pics/W7N6LPp0SmY12986125.jpg'}, {'end': 13367.779, 'src': 'embed', 'start': 13341.745, 'weight': 2, 'content': [{'end': 13349.214, 'text': 'You are totally correct that the model has now seen 55, 000 different permutation combinations, but we are not sure right?', 'start': 13341.745, 'duration': 7.469}, {'end': 13352.578, 'text': 'The data set might have a duplicate image also.', 'start': 13349.474, 'duration': 3.104}, {'end': 13354.881, 'text': 'I mean, they might have a duplicate five written also.', 'start': 13352.678, 'duration': 2.203}, {'end': 13360.237, 'text': 'So five might be, might be written multiple times by a same person, so he might have written it.', 'start': 13354.901, 'duration': 5.336}, {'end': 13367.779, 'text': "but yes, you're totally right that we assume that our assumption is that the data sets are totally unique in nature and if we plot them,", 'start': 13360.237, 'duration': 7.542}], 'summary': 'Model trained on 55,000 permutations, but uncertainty due to possible duplicate images and repeated data points.', 'duration': 26.034, 'max_score': 13341.745, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/W7N6LPp0SmY/pics/W7N6LPp0SmY13341745.jpg'}, {'end': 13904.471, 'src': 'embed', 'start': 13849.099, 'weight': 3, 'content': [{'end': 13850.922, 'text': 'So, till last class, what we were doing?', 'start': 13849.099, 'duration': 1.823}, {'end': 13855.689, 'text': 'We were doing Jupiter notebooks right?', 'start': 13850.942, 'duration': 4.747}, {'end': 13859.173, 'text': 'But this time we will be doing Jupiter lab.', 'start': 13855.849, 'duration': 3.324}, {'end': 13865.042, 'text': 'Now you will see this is a new version of Jupiter notebook that will open up in your screens.', 'start': 13860.135, 'duration': 4.907}, {'end': 13898.088, 'text': 'So now you can see that this is a new interface that is open for you all.', 'start': 13894.807, 'duration': 3.281}, {'end': 13904.471, 'text': 'All the people who have laptop can also do this right now with us only so that they can get the idea.', 'start': 13899.889, 'duration': 4.582}], 'summary': 'Transitioning from jupyter notebooks to jupyter lab for a new interface and enhanced functionality.', 'duration': 55.372, 'max_score': 13849.099, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/W7N6LPp0SmY/pics/W7N6LPp0SmY13849099.jpg'}, {'end': 14481.54, 'src': 'embed', 'start': 14448.33, 'weight': 4, 'content': [{'end': 14449.451, 'text': "it's called labels, right?", 'start': 14448.33, 'duration': 1.121}, {'end': 14452.947, 'text': 'So we got the data set and then we started doing it on Keras.', 'start': 14449.804, 'duration': 3.143}, {'end': 14454.889, 'text': 'But today we will do it on TensorFlow also.', 'start': 14453.027, 'duration': 1.862}, {'end': 14455.99, 'text': 'Why on TensorFlow?', 'start': 14455.109, 'duration': 0.881}, {'end': 14464.297, 'text': 'Because I want to give you a basic idea how code in TensorFlow is also written and how Keras is helping you in you know how much lowering the works.', 'start': 14456.01, 'duration': 8.287}, {'end': 14481.54, 'text': "Otherwise, you won't get to know that how much things Keras is doing in on itself, right? I'll do the first basic.", 'start': 14464.417, 'duration': 17.123}], 'summary': 'Introduction to using keras and tensorflow for data labeling and basic coding.', 'duration': 33.21, 'max_score': 14448.33, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/W7N6LPp0SmY/pics/W7N6LPp0SmY14448330.jpg'}], 'start': 12439.681, 'title': 'Deep learning basics and tools', 'summary': 'Discusses binary vs multi-class classification, building a basic neural network with low accuracy, permutations, combinations, jupiter lab introduction, and tensorflow emulation basics.', 'chapters': [{'end': 12648.423, 'start': 12439.681, 'title': 'Deep learning: binary vs multi-class classification', 'summary': 'Discusses the distinction between binary and multi-class classification in deep learning, emphasizing the use of softmax activation function to calculate probabilities, enabling comparison and decision-making in multi-class scenarios.', 'duration': 208.742, 'highlights': ['The softmax activation function ensures that the sum of probabilities for all classes always equals one, facilitating comparison and decision-making in multi-class classification scenarios.', 'In multi-class classification, the model provides probabilities for each class, and the highest probability determines the predicted class, enabling decision-making based on the calculated probabilities.', 'Deep learning involves utilizing the softmax activation function to calculate probabilities for different classes, aiding in decision-making for multi-class classification tasks.']}, {'end': 13318.38, 'start': 12648.683, 'title': 'Building basic neural network', 'summary': 'Covers the process of building a basic neural network using keras from inside tensorflow, with a demonstration of model compilation and training, resulting in a very low accuracy of 13%, highlighting the importance of understanding the basics before delving into deep learning.', 'duration': 669.697, 'highlights': ['The process of building a basic neural network using Keras from inside TensorFlow, including model compilation and training, results in a very low accuracy of 13%, emphasizing the significance of understanding the basics before diving into deep learning.', 'The demonstration of model compilation and training, where the model is trained for 20 epochs, showcases the iterative process of feed forward and backpropagation in deep learning.', "Explanation of the model's architecture, including the use of a categorical cross entropy loss function and the stochastic gradient descent (SGD) optimizer, provides insights into the various components involved in training a neural network.", 'The instructor emphasizes the importance of understanding basics and gradually learning about sequential modeling, dense layers, different types of losses (e.g., categorical cross entropy), and optimizers (e.g., SGD, Adam) in the context of building neural networks from scratch.']}, {'end': 13716.813, 'start': 13318.38, 'title': 'Understanding permutations and combinations', 'summary': "Discusses the model's exposure to 55,000 unique permutations and combinations, with 5,500 images per category, and suggests reading a deep learning book with freely available resources.", 'duration': 398.433, 'highlights': ['The model has now seen 55,000 different permutation combinations of data available to it, with 5,500 images per category.', 'The recommended deep learning book and freely available resources provide valuable learning material for understanding deep learning concepts and applications.', 'The chapter reinforces the importance of reading and understanding a deep learning book to gain insights and knowledge about deep learning concepts and applications.']}, {'end': 14250.653, 'start': 13721.482, 'title': 'Introduction to jupiter lab', 'summary': 'Introduces the audience to jupiter lab, emphasizing its features and capabilities, including navigating to the folder, creating a new notebook, accessing python console and cmd, and uploading data files on jupyterlab.', 'duration': 529.171, 'highlights': ['Jupiter Lab is introduced as a new version of Jupiter notebook, emphasizing its new interface and capabilities for creating notebooks and accessing Python console and CMD.', "Instructions are provided on creating a new folder, renaming it, and creating a Jupiter notebook within the folder, with a specific example of creating a folder named 'tensorflow' and a notebook within it.", 'Features of Jupyter Lab are highlighted, including the ability to open tabs, access cell tools, and perform commands such as opening new consoles and closing files, as well as the capability to upload data files while running on a server instance.']}, {'end': 14834.99, 'start': 14254.648, 'title': 'Tensorflow emulation basics', 'summary': "Covers the basics of tensorflow emulation and includes a demonstration of working with the 'eminist' dataset, emphasizing the comparison between tensorflow and keras, with an encouragement for practical application and revision.", 'duration': 580.342, 'highlights': ['The chapter emphasizes the comparison between TensorFlow and Keras, with an encouragement for practical application and revision.', "A demonstration of working with the 'eminist' dataset, including downloading, one-hot encoding, and handling warnings.", 'Introduction to TensorFlow emulation basics, including imports, data retrieval, and version checking.']}], 'duration': 2395.309, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/W7N6LPp0SmY/pics/W7N6LPp0SmY12439681.jpg', 'highlights': ['The softmax activation function aids in decision-making for multi-class classification tasks.', 'Understanding the basics before diving into deep learning is emphasized.', 'The model has seen 55,000 different permutation combinations of data.', 'Jupyter Lab is introduced as a new version of Jupiter notebook.', 'The chapter emphasizes the comparison between TensorFlow and Keras.']}, {'end': 17059.305, 'segs': [{'end': 14863.673, 'src': 'embed', 'start': 14834.99, 'weight': 1, 'content': [{'end': 14838.193, 'text': "now let's see what is inside mnist if we check the type of this.", 'start': 14834.99, 'duration': 3.203}, {'end': 14844.897, 'text': "So it says it's a dataset type of a thing.", 'start': 14842.475, 'duration': 2.422}, {'end': 14847.359, 'text': "Let's see what things it can help us.", 'start': 14845.137, 'duration': 2.222}, {'end': 14851.503, 'text': 'Okay, so it has actually three things, test, train, and validation.', 'start': 14847.98, 'duration': 3.523}, {'end': 14853.785, 'text': "Let's go inside what is this test train.", 'start': 14852.003, 'duration': 1.782}, {'end': 14855.386, 'text': 'And you can see that there is .', 'start': 14854.105, 'duration': 1.281}, {'end': 14856.087, 'text': 'index and .', 'start': 14855.386, 'duration': 0.701}, {'end': 14857.748, 'text': "count It means it's actually a tuple.", 'start': 14856.087, 'duration': 1.661}, {'end': 14863.673, 'text': "It's actually inherited from a tuple thing, right? Let's go inside train.", 'start': 14858.709, 'duration': 4.964}], 'summary': "The mnist dataset includes three main components: test, train, and validation, with 'train' being inherited from a tuple.", 'duration': 28.683, 'max_score': 14834.99, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/W7N6LPp0SmY/pics/W7N6LPp0SmY14834990.jpg'}, {'end': 14965.969, 'src': 'embed', 'start': 14911.799, 'weight': 0, 'content': [{'end': 14914.42, 'text': 'so what you have to do is that you have to do a reshape of this.', 'start': 14911.799, 'duration': 2.621}, {'end': 14923.525, 'text': 'also, we know 784 is a square of 28 cross 28, and we know these are the images of 28 plus 28.', 'start': 14914.42, 'duration': 9.105}, {'end': 14929.127, 'text': 'now you get 28 cross 28..', 'start': 14923.525, 'duration': 5.602}, {'end': 14929.827, 'text': 'we will do a one.', 'start': 14929.127, 'duration': 0.7}, {'end': 14930.347, 'text': 'also why?', 'start': 14929.827, 'duration': 0.52}, {'end': 14934.349, 'text': 'because i want to show that this is a one channel image.', 'start': 14930.347, 'duration': 4.002}, {'end': 14965.969, 'text': 'now, if i do a plot, wait, what we can do more is that we can add it.', 'start': 14934.349, 'duration': 31.62}], 'summary': 'Reshape the 784 images into 28x28 and convert to one-channel images.', 'duration': 54.17, 'max_score': 14911.799, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/W7N6LPp0SmY/pics/W7N6LPp0SmY14911799.jpg'}, {'end': 15141.038, 'src': 'embed', 'start': 15118.232, 'weight': 2, 'content': [{'end': 15126.431, 'text': 'just try to plot it now, because if you see the shape of mns train images, it is actually 784, comma 0..', 'start': 15118.232, 'duration': 8.199}, {'end': 15129.372, 'text': 'Now, our image is not actually a 784 comma 0, right?', 'start': 15126.431, 'duration': 2.941}, {'end': 15133.454, 'text': 'It is flattened image, just to save the data, right?', 'start': 15129.572, 'duration': 3.882}, {'end': 15137.096, 'text': "So what we're going to do is that we did a reshape of this image.", 'start': 15133.995, 'duration': 3.101}, {'end': 15141.038, 'text': 'Because we already know from the Yanlikun site.', 'start': 15138.757, 'duration': 2.281}], 'summary': 'Data reshaped from 784 to flattened image', 'duration': 22.806, 'max_score': 15118.232, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/W7N6LPp0SmY/pics/W7N6LPp0SmY15118232.jpg'}, {'end': 15349.966, 'src': 'embed', 'start': 15262.844, 'weight': 5, 'content': [{'end': 15263.545, 'text': 'So what we are going to do?', 'start': 15262.844, 'duration': 0.701}, {'end': 15265.826, 'text': 'we are going to say tf.placeholder.', 'start': 15263.545, 'duration': 2.281}, {'end': 15267.908, 'text': 'So first is actually the dtype.', 'start': 15266.287, 'duration': 1.621}, {'end': 15270.87, 'text': 'dtype is df.float32.', 'start': 15267.948, 'duration': 2.922}, {'end': 15278.516, 'text': 'Correct Then it asks for the second thing is that is shape.', 'start': 15272.892, 'duration': 5.624}, {'end': 15279.437, 'text': 'Shape is equal to none.', 'start': 15278.576, 'duration': 0.861}, {'end': 15280.277, 'text': 'No, no, no, no.', 'start': 15279.517, 'duration': 0.76}, {'end': 15285.882, 'text': "We don't give shape is equal to 28 cross 28.", 'start': 15281.118, 'duration': 4.764}, {'end': 15286.342, 'text': 'We give it.', 'start': 15285.882, 'duration': 0.46}, {'end': 15296.892, 'text': 'We give it the full image.', 'start': 15295.771, 'duration': 1.121}, {'end': 15300.374, 'text': 'We are not going to do the unfolding of the image.', 'start': 15296.992, 'duration': 3.382}, {'end': 15301.975, 'text': 'We will let it remain flatten only.', 'start': 15300.394, 'duration': 1.581}, {'end': 15307.458, 'text': 'Right Now we are going to make the weights.', 'start': 15302.255, 'duration': 5.203}, {'end': 15311.48, 'text': 'How do we make the weights? Weights is equal to tf.variable.', 'start': 15308.338, 'duration': 3.142}, {'end': 15318.164, 'text': 'Right By default we are going to initialize them with zeros.', 'start': 15313.221, 'duration': 4.943}, {'end': 15322.847, 'text': 'Correct. How many zeros??', 'start': 15320.906, 'duration': 1.941}, {'end': 15330.622, 'text': '784, the number of inputs and the next column output will be 100, only right.', 'start': 15324.14, 'duration': 6.482}, {'end': 15331.702, 'text': 'can anybody explain me?', 'start': 15330.622, 'duration': 1.08}, {'end': 15347.125, 'text': 'how did i came with this number of weight, 784 into 100?', 'start': 15331.702, 'duration': 15.423}, {'end': 15349.966, 'text': "anybody? no, i'll tell you.", 'start': 15347.125, 'duration': 2.841}], 'summary': 'Creating placeholders for image data, initializing weights for neural network with 784 inputs and 100 outputs.', 'duration': 87.122, 'max_score': 15262.844, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/W7N6LPp0SmY/pics/W7N6LPp0SmY15262844.jpg'}, {'end': 15664.237, 'src': 'embed', 'start': 15623.91, 'weight': 7, 'content': [{'end': 15626.091, 'text': 'Everything is connected to everything right?', 'start': 15623.91, 'duration': 2.181}, {'end': 15629.033, 'text': 'So what you can do is how many weights will be there?', 'start': 15626.551, 'duration': 2.482}, {'end': 15631.834, 'text': 'Input into number of output, right?', 'start': 15629.153, 'duration': 2.681}, {'end': 15635.375, 'text': 'Input into number of output is equal to the number of weights.', 'start': 15632.014, 'duration': 3.361}, {'end': 15640.578, 'text': 'But how is a bias connected? A bias is connected somewhat like this.', 'start': 15635.856, 'duration': 4.722}, {'end': 15642.519, 'text': 'Let me give you an example here.', 'start': 15640.618, 'duration': 1.901}, {'end': 15646.84, 'text': 'A bias is connected to every output neuron like this.', 'start': 15643.559, 'duration': 3.281}, {'end': 15647.921, 'text': 'B1, then B2.', 'start': 15647.581, 'duration': 0.34}, {'end': 15664.237, 'text': 'then b3, b4, or you just imagine, just imagine, that bias is nothing, but bias is actually a input only, but that input value is one.', 'start': 15651.687, 'duration': 12.55}], 'summary': 'Neural network connections explained with example and bias as an input with value one.', 'duration': 40.327, 'max_score': 15623.91, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/W7N6LPp0SmY/pics/W7N6LPp0SmY15623910.jpg'}, {'end': 15832.808, 'src': 'embed', 'start': 15787.111, 'weight': 8, 'content': [{'end': 15788.752, 'text': 'I have to read a little bit more.', 'start': 15787.111, 'duration': 1.641}, {'end': 15790.253, 'text': 'Okay, okay, okay.', 'start': 15788.992, 'duration': 1.261}, {'end': 15800.865, 'text': 'I think I suggested you guys a book called Neural Networks and Deep Learning by Michael Nielsen.', 'start': 15793.283, 'duration': 7.582}, {'end': 15805.227, 'text': 'That is a very, very genuine book.', 'start': 15802.546, 'duration': 2.681}, {'end': 15806.487, 'text': 'Please read that book.', 'start': 15805.427, 'duration': 1.06}, {'end': 15807.667, 'text': 'That is very helpful.', 'start': 15806.527, 'duration': 1.14}, {'end': 15814.59, 'text': 'I was reading that bias is threshold value.', 'start': 15811.629, 'duration': 2.961}, {'end': 15822.152, 'text': 'Threshold value of what? I did not get it.', 'start': 15816.29, 'duration': 5.862}, {'end': 15832.808, 'text': 'okay, okay, you are saying that I did not get the book.', 'start': 15828.706, 'duration': 4.102}], 'summary': "Suggested reading 'neural networks and deep learning' by michael nielsen; emphasized its helpfulness.", 'duration': 45.697, 'max_score': 15787.111, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/W7N6LPp0SmY/pics/W7N6LPp0SmY15787111.jpg'}, {'end': 16002.73, 'src': 'embed', 'start': 15945.734, 'weight': 9, 'content': [{'end': 15952.441, 'text': "So a bias is actually a safe, safe side prediction, where we're not a prediction but a safe side calculation.", 'start': 15945.734, 'duration': 6.707}, {'end': 15954.824, 'text': 'that we do to avoid things.', 'start': 15952.441, 'duration': 2.383}, {'end': 15964.695, 'text': 'So whenever you will do a binary classification you have only one thing right either true or false right.', 'start': 15959.629, 'duration': 5.066}, {'end': 15974.052, 'text': 'at that time, when you calculate a cost function, you will get a error there that how will you produce for zero and how will you produce for one?', 'start': 15967.029, 'duration': 7.023}, {'end': 15977.573, 'text': "that's why we need to introduce this thing called bias instead.", 'start': 15974.052, 'duration': 3.521}, {'end': 15991.058, 'text': 'so just consider bias as another weight where it is actually connected to a input where input value is equal to one weight or weighted summation of no,', 'start': 15977.573, 'duration': 13.485}, {'end': 15992.379, 'text': 'no, no, no, no.', 'start': 15991.058, 'duration': 1.321}, {'end': 15994.759, 'text': 'weighted only weight, only weight.', 'start': 15992.379, 'duration': 2.38}, {'end': 16002.73, 'text': "now i'm saying just consider, just consider, just for simplifying the process, just consider bias as a weight only.", 'start': 15994.759, 'duration': 7.971}], 'summary': 'Introducing bias in binary classification for error reduction.', 'duration': 56.996, 'max_score': 15945.734, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/W7N6LPp0SmY/pics/W7N6LPp0SmY15945734.jpg'}, {'end': 16088.411, 'src': 'embed', 'start': 16048.244, 'weight': 12, 'content': [{'end': 16051.689, 'text': 'that was just single neuron And affine was happening.', 'start': 16048.244, 'duration': 3.445}, {'end': 16059.398, 'text': 'Remember? I was just showing you a single neuron where there is only one circle and affine is going on and activation is going on.', 'start': 16052.029, 'duration': 7.369}, {'end': 16066.968, 'text': 'That is what represents a single neuron entity of a neural network.', 'start': 16062.162, 'duration': 4.806}, {'end': 16086.731, 'text': "Now let's say that we have declared these three things x, w and b.", 'start': 16082.83, 'duration': 3.901}, {'end': 16088.411, 'text': 'These are the three things that we want to declare.', 'start': 16086.731, 'duration': 1.68}], 'summary': 'A single neuron with affine transformation and activation represents a neural network, with x, w, and b as the key components.', 'duration': 40.167, 'max_score': 16048.244, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/W7N6LPp0SmY/pics/W7N6LPp0SmY16048244.jpg'}, {'end': 16402.312, 'src': 'embed', 'start': 16365.686, 'weight': 13, 'content': [{'end': 16372.407, 'text': 'all the probabilities that it is giving you out for all the outputs, the sum of all this should be equal to 100 or 1..', 'start': 16365.686, 'duration': 6.721}, {'end': 16377.445, 'text': "I mean If you're taking in terms of percentage, it should be 100.", 'start': 16372.407, 'duration': 5.038}, {'end': 16381.986, 'text': "And if you're taking in terms of probability, it should be 1.", 'start': 16377.445, 'duration': 4.541}, {'end': 16401.251, 'text': 'Make sense? Sanjay? Omar, making sense to you also, right? Helping out? Cool.', 'start': 16381.986, 'duration': 19.265}, {'end': 16402.312, 'text': 'Cool, cool, cool.', 'start': 16401.67, 'duration': 0.642}], 'summary': 'Probabilities should sum to 100% or 1 in terms of percentage or probability, respectively.', 'duration': 36.626, 'max_score': 16365.686, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/W7N6LPp0SmY/pics/W7N6LPp0SmY16365686.jpg'}, {'end': 16458.874, 'src': 'embed', 'start': 16434.115, 'weight': 15, 'content': [{'end': 16443.162, 'text': 'If you want, I can send you all the sheets with all these formulas and all, right? So we asked, first we did the affine.', 'start': 16434.115, 'duration': 9.047}, {'end': 16447.125, 'text': 'We did a matrix multiplication of x and w.', 'start': 16443.742, 'duration': 3.383}, {'end': 16448.385, 'text': 'Then we added bias into it.', 'start': 16447.125, 'duration': 1.26}, {'end': 16456.572, 'text': 'And then we did a softmax of the whole thing, right? We have done this thing, right? This is the one that we are going to calculate.', 'start': 16448.666, 'duration': 7.906}, {'end': 16458.874, 'text': 'This is the y that is being calculated.', 'start': 16456.693, 'duration': 2.181}], 'summary': 'Performed affine transformation, matrix multiplication, bias addition, and softmax calculation for y.', 'duration': 24.759, 'max_score': 16434.115, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/W7N6LPp0SmY/pics/W7N6LPp0SmY16434115.jpg'}, {'end': 16529.07, 'src': 'embed', 'start': 16497.193, 'weight': 14, 'content': [{'end': 16498.813, 'text': 'why it will be 10?', 'start': 16497.193, 'duration': 1.62}, {'end': 16501.116, 'text': 'because our image will have only one thing right.', 'start': 16498.813, 'duration': 2.303}, {'end': 16506.861, 'text': 'if this is the image, the output will be 4 right.', 'start': 16501.116, 'duration': 5.745}, {'end': 16512.965, 'text': 'why i am giving 10 here?', 'start': 16506.861, 'duration': 6.104}, {'end': 16515.226, 'text': 'because we did a one hot encoding.', 'start': 16512.965, 'duration': 2.261}, {'end': 16517.469, 'text': 'you remember on the above.', 'start': 16515.226, 'duration': 2.243}, {'end': 16519.871, 'text': 'we did this one hot encoding here at this point.', 'start': 16517.469, 'duration': 2.402}, {'end': 16523.205, 'text': 'Now, what does one hot encoding does?', 'start': 16521.262, 'duration': 1.943}, {'end': 16529.07, 'text': 'one hot encoding says that your data is actually not in the form of one, two, three, four.', 'start': 16523.205, 'duration': 5.865}], 'summary': 'One hot encoding resulted in output of 10 instead of 4 due to presence of one thing in the image.', 'duration': 31.877, 'max_score': 16497.193, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/W7N6LPp0SmY/pics/W7N6LPp0SmY16497193.jpg'}, {'end': 16673.989, 'src': 'embed', 'start': 16645.115, 'weight': 16, 'content': [{'end': 16651.56, 'text': "okay, I'll tell you what is happening here is, at this particular step, what we are trying to do.", 'start': 16645.115, 'duration': 6.445}, {'end': 16654.843, 'text': 'we are trying to calculate the errors for back proposition.', 'start': 16651.56, 'duration': 3.283}, {'end': 16656.585, 'text': 'What way we are going to go?', 'start': 16655.765, 'duration': 0.82}, {'end': 16658.346, 'text': 'We are trying to calculate the errors.', 'start': 16656.625, 'duration': 1.721}, {'end': 16664.427, 'text': 'And you remember error was actually what? In the last class we discussed this also.', 'start': 16660.026, 'duration': 4.401}, {'end': 16673.989, 'text': 'What was error? Error was nothing but y actual minus y predicted.', 'start': 16664.886, 'duration': 9.103}], 'summary': 'Calculating errors for backpropagation to optimize y predicted', 'duration': 28.874, 'max_score': 16645.115, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/W7N6LPp0SmY/pics/W7N6LPp0SmY16645115.jpg'}, {'end': 16761.778, 'src': 'embed', 'start': 16733.34, 'weight': 17, 'content': [{'end': 16734.919, 'text': 'And I have reduced it to the indexes.', 'start': 16733.34, 'duration': 1.579}, {'end': 16735.619, 'text': 'First index.', 'start': 16735, 'duration': 0.619}, {'end': 16738.401, 'text': 'Right And then I took out the mean of it.', 'start': 16736.36, 'duration': 2.041}, {'end': 16741.221, 'text': 'Because it is the cross entropy is the mean actually.', 'start': 16738.481, 'duration': 2.74}, {'end': 16743.122, 'text': 'To divide by the number of observation.', 'start': 16741.721, 'duration': 1.401}, {'end': 16744.362, 'text': 'That is called the cross entropy.', 'start': 16743.142, 'duration': 1.22}, {'end': 16746.401, 'text': 'So you have the cross entropy with you now.', 'start': 16744.682, 'duration': 1.719}, {'end': 16749.863, 'text': 'Everybody gets the idea.', 'start': 16748.723, 'duration': 1.14}, {'end': 16758.865, 'text': 'I will take a pause here.', 'start': 16749.883, 'duration': 8.982}, {'end': 16761.778, 'text': 'have a look at the screen.', 'start': 16759.577, 'duration': 2.201}], 'summary': 'Reduced indexes, calculated cross entropy, and explained its significance.', 'duration': 28.438, 'max_score': 16733.34, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/W7N6LPp0SmY/pics/W7N6LPp0SmY16733340.jpg'}, {'end': 17041.689, 'src': 'embed', 'start': 17015.69, 'weight': 18, 'content': [{'end': 17022.033, 'text': "Let's say it's interactive session, right? It's actually a session only, but you can handle it very carefully.", 'start': 17015.69, 'duration': 6.343}, {'end': 17031.841, 'text': 'right. so it actually made for what you call for all your jupiter notebooks, so you can do a interactive session rather than doing tf dot session.', 'start': 17022.554, 'duration': 9.287}, {'end': 17036.405, 'text': 'you can do interactive session right now.', 'start': 17031.841, 'duration': 4.564}, {'end': 17041.689, 'text': "what you're going to do first thing, when you sorry, yeah, what is between the session and the interactive session?", 'start': 17036.405, 'duration': 5.284}], 'summary': 'The session focuses on transitioning from regular to interactive sessions in jupyter notebooks.', 'duration': 25.999, 'max_score': 17015.69, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/W7N6LPp0SmY/pics/W7N6LPp0SmY17015690.jpg'}], 'start': 14834.99, 'title': 'Understanding neural networks concepts', 'summary': 'Explores the mnist dataset with test, train, and validation data, demonstrates tensorflow variables and placeholders creation, emphasizes the importance of bias in neural networks, discusses the softmax function, and explains cross entropy in neural networks.', 'chapters': [{'end': 15171.619, 'start': 14834.99, 'title': 'Understanding mnist dataset', 'summary': 'Explores the mnist dataset, which contains test, train, and validation data, with the train data consisting of images and labels, and demonstrates how to reshape and plot the images, along with explaining the function of the dataset.', 'duration': 336.629, 'highlights': ['The MNIST dataset contains test, train, and validation data, with the train data including images and labels. This provides an overview of the contents of the MNIST dataset.', 'The train data set consists of images and labels, with each image being a flattened array of 784 elements. This highlights the structure of the train data set and the flattened nature of the images.', 'The process of reshaping the flattened image data into 28x28 dimensions is explained, aligning with the standard image size in the MNIST dataset. This emphasizes the importance of reshaping the flattened image data to match the standard image size in the MNIST dataset.', 'Demonstration of plotting the images from the MNIST dataset after reshaping and converting them to grayscale. This showcases the practical process of plotting the images from the MNIST dataset after necessary modifications.', 'Explanation of the function of the MNIST dataset, including its provision of images and corresponding labels for machine learning tasks. This provides a clear understanding of the purpose and function of the MNIST dataset in machine learning applications.']}, {'end': 15780.346, 'start': 15171.619, 'title': 'Tensorflow variables and placeholder', 'summary': 'Covers the creation of tensorflow variables and placeholders, including the process of making placeholders for images and labels, initializing weights and biases, and understanding the number of weights and biases required for the neural network.', 'duration': 608.727, 'highlights': ['Creation of TensorFlow placeholders for images and labels The process of making placeholders for images and labels involves defining the data type as df.float32 and specifying the shape as the full image (28x28), without unfolding it.', 'Initialization of weights with zeros and determination of the number of weights required Weights are initialized using tf.variable with zeros, and the number of weights is determined as 784x100, corresponding to the inputs and output columns.', 'Understanding the number of biases required based on the neural network structure The explanation involves understanding that the number of biases is equal to the number of output neurons, which in this case is 10, and that biases connect to every output neuron.']}, {'end': 16192.841, 'start': 15787.111, 'title': 'Understanding bias in neural networks', 'summary': "Emphasizes the importance of reading the book 'neural networks and deep learning' by michael nielsen. it delves into the concept of bias in neural networks, explaining its role in simplifying mathematical calculations and ensuring accurate outputs in binary classification problems.", 'duration': 405.73, 'highlights': ["The importance of reading 'Neural Networks and Deep Learning' by Michael Nielsen is emphasized. The book is recommended as very helpful and essential for those interested in delving into the theory of neural networks and deep learning.", 'Explanation of the role of bias in simplifying mathematical calculations and ensuring accurate outputs in binary classification problems. Bias is described as a threshold value that simplifies mathematical calculations and ensures accurate outputs in binary classification problems, where there are only two possible outcomes: true or false.', 'The concept of bias as a safe side calculation to avoid errors in binary classification problems is discussed. Bias is explained as a safe side calculation to avoid errors in binary classification problems, ensuring accurate outputs for both true and false outcomes.', 'The use of bias as another weight in simplifying the process is clarified. Bias is compared to another weight and considered as a simplification in the process, particularly in the affine equation and the perceptron model.', 'The role of bias in the single neuron entity of a neural network is explained. Bias is detailed as a key component in the representation of a single neuron entity in a neural network, particularly in the context of the affine and activation processes.']}, {'end': 16560.46, 'start': 16192.841, 'title': 'Understanding softmax function', 'summary': 'Discusses the working of the softmax function in neural networks, explaining how it determines the probabilities of different classes and ensures a total probability of 1, with an emphasis on the one-hot encoding of data.', 'duration': 367.619, 'highlights': ['The softmax function determines the probabilities of different classes and ensures the sum of all these probabilities equals 1, or 100% if represented as percentages.', 'Explanation of how the softmax function converts data into a probability distribution, with an emphasis on one-hot encoding and the resulting 10-length output for each image.', 'The process of matrix multiplication and addition of biases before applying the softmax function to calculate the final output in neural networks.']}, {'end': 17059.305, 'start': 16560.46, 'title': 'Understanding cross entropy in neural networks', 'summary': 'Explains the concept of cross entropy in neural networks, which is used to calculate errors for back propagation, minimize the cross entropy using gradient descent, and run the process in an interactive session in tensorflow.', 'duration': 498.845, 'highlights': ['The process involves calculating the errors for back propagation by subtracting the actual values from the predicted values, then summing and taking the mean of the results to obtain the cross entropy, which is used to minimize the error in the neural network through gradient descent.', 'The concept of cross entropy is demonstrated using the example of image recognition, where the actual and predicted values are compared and manipulated to derive the cross entropy for multiple images within a neural network.', 'The use of an interactive session in TensorFlow is explained, highlighting its advantages in handling the training process and automatic closure after completion, particularly suitable for Jupyter notebooks.']}], 'duration': 2224.315, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/W7N6LPp0SmY/pics/W7N6LPp0SmY14834990.jpg', 'highlights': ['The process of reshaping the flattened image data into 28x28 dimensions is explained, aligning with the standard image size in the MNIST dataset.', 'The MNIST dataset contains test, train, and validation data, with the train data including images and labels. This provides an overview of the contents of the MNIST dataset.', 'The train data set consists of images and labels, with each image being a flattened array of 784 elements. This highlights the structure of the train data set and the flattened nature of the images.', 'Demonstration of plotting the images from the MNIST dataset after reshaping and converting them to grayscale. This showcases the practical process of plotting the images from the MNIST dataset after necessary modifications.', 'Explanation of the function of the MNIST dataset, including its provision of images and corresponding labels for machine learning tasks. This provides a clear understanding of the purpose and function of the MNIST dataset in machine learning applications.', 'Creation of TensorFlow placeholders for images and labels The process of making placeholders for images and labels involves defining the data type as df.float32 and specifying the shape as the full image (28x28), without unfolding it.', 'Initialization of weights with zeros and determination of the number of weights required Weights are initialized using tf.variable with zeros, and the number of weights is determined as 784x100, corresponding to the inputs and output columns.', 'Understanding the number of biases required based on the neural network structure The explanation involves understanding that the number of biases is equal to the number of output neurons, which in this case is 10, and that biases connect to every output neuron.', "The importance of reading 'Neural Networks and Deep Learning' by Michael Nielsen is emphasized. The book is recommended as very helpful and essential for those interested in delving into the theory of neural networks and deep learning.", 'Explanation of the role of bias in simplifying mathematical calculations and ensuring accurate outputs in binary classification problems. Bias is described as a threshold value that simplifies mathematical calculations and ensures accurate outputs in binary classification problems, where there are only two possible outcomes: true or false.', 'The concept of bias as a safe side calculation to avoid errors in binary classification problems is discussed. Bias is explained as a safe side calculation to avoid errors in binary classification problems, ensuring accurate outputs for both true and false outcomes.', 'The use of bias as another weight in simplifying the process is clarified. Bias is compared to another weight and considered as a simplification in the process, particularly in the affine equation and the perceptron model.', 'The role of bias in the single neuron entity of a neural network is explained. Bias is detailed as a key component in the representation of a single neuron entity in a neural network, particularly in the context of the affine and activation processes.', 'The softmax function determines the probabilities of different classes and ensures the sum of all these probabilities equals 1, or 100% if represented as percentages.', 'Explanation of how the softmax function converts data into a probability distribution, with an emphasis on one-hot encoding and the resulting 10-length output for each image.', 'The process of matrix multiplication and addition of biases before applying the softmax function to calculate the final output in neural networks.', 'The process involves calculating the errors for back propagation by subtracting the actual values from the predicted values, then summing and taking the mean of the results to obtain the cross entropy, which is used to minimize the error in the neural network through gradient descent.', 'The concept of cross entropy is demonstrated using the example of image recognition, where the actual and predicted values are compared and manipulated to derive the cross entropy for multiple images within a neural network.', 'The use of an interactive session in TensorFlow is explained, highlighting its advantages in handling the training process and automatic closure after completion, particularly suitable for Jupyter notebooks.']}, {'end': 18245.416, 'segs': [{'end': 17092.78, 'src': 'embed', 'start': 17059.305, 'weight': 2, 'content': [{'end': 17067.411, 'text': 'it is meant for that only right, once you make the, once you make the session, what is the first thing you always do?', 'start': 17059.305, 'duration': 8.106}, {'end': 17080.589, 'text': 'you will always remember global variable initialization.', 'start': 17067.411, 'duration': 13.178}, {'end': 17084.232, 'text': 'there are three things placeholders, variables and constant.', 'start': 17080.589, 'duration': 3.643}, {'end': 17092.78, 'text': 'you can run placeholders and constant directly inside a session, but if you want to run any variable inside your session, then what you have to do?', 'start': 17084.232, 'duration': 8.548}], 'summary': 'Training session emphasizes global variable initialization and different types of values.', 'duration': 33.475, 'max_score': 17059.305, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/W7N6LPp0SmY/pics/W7N6LPp0SmY17059305.jpg'}, {'end': 17228.86, 'src': 'embed', 'start': 17194.51, 'weight': 0, 'content': [{'end': 17197.792, 'text': '550 times because I have 55, 000 images right?', 'start': 17194.51, 'duration': 3.282}, {'end': 17201.915, 'text': 'Making sense to everyone right?', 'start': 17200.614, 'duration': 1.301}, {'end': 17205.678, 'text': 'Because we have 55, 000 images and I am taking only 100 images at a time.', 'start': 17202.075, 'duration': 3.603}, {'end': 17210.701, 'text': 'So I have to take 550 times.', 'start': 17206.178, 'duration': 4.523}, {'end': 17215.004, 'text': 'I have to run this thing to complete all the images right?', 'start': 17210.701, 'duration': 4.303}, {'end': 17216.145, 'text': 'So what is happening now?', 'start': 17215.344, 'duration': 0.801}, {'end': 17218.637, 'text': 'It will give me batch X and batch Y.', 'start': 17216.676, 'duration': 1.961}, {'end': 17228.86, 'text': "Batch X will contain all the images and batch Y will contain the correspondingly labels of the thing, right? I'll do nothing.", 'start': 17218.637, 'duration': 10.223}], 'summary': '550 iterations needed to process 55,000 images in batches of 100.', 'duration': 34.35, 'max_score': 17194.51, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/W7N6LPp0SmY/pics/W7N6LPp0SmY17194510.jpg'}, {'end': 17332.44, 'src': 'embed', 'start': 17304.087, 'weight': 1, 'content': [{'end': 17308.789, 'text': "So he'll say, okay, Y is actually a calculation of X, W, and B.", 'start': 17304.087, 'duration': 4.702}, {'end': 17311.13, 'text': 'It will ask, what is X? X is again a placeholder.', 'start': 17308.789, 'duration': 2.341}, {'end': 17313.691, 'text': 'You get the value from your feed date.', 'start': 17311.15, 'duration': 2.541}, {'end': 17316.853, 'text': 'But who are W and B? It will say, W and B are here.', 'start': 17314.112, 'duration': 2.741}, {'end': 17326.417, 'text': 'Everybody gets the idea how a session things work in TensorFlow.', 'start': 17321.215, 'duration': 5.202}, {'end': 17330.139, 'text': 'Please ask your questions right now.', 'start': 17328.198, 'duration': 1.941}, {'end': 17332.44, 'text': 'Otherwise, it will become a little messy afterwards.', 'start': 17330.159, 'duration': 2.281}], 'summary': 'Explanation of y as a calculation of x, w, and b in tensorflow session.', 'duration': 28.353, 'max_score': 17304.087, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/W7N6LPp0SmY/pics/W7N6LPp0SmY17304087.jpg'}, {'end': 17417.535, 'src': 'embed', 'start': 17391.842, 'weight': 8, 'content': [{'end': 17397.741, 'text': "okay, The concept of batches came because we don't have as much as RAM we can afford right?", 'start': 17391.842, 'duration': 5.899}, {'end': 17401.283, 'text': 'Because we have a limited RAM resource right?', 'start': 17398.041, 'duration': 3.242}, {'end': 17402.984, 'text': "Let's say we have 1 million images.", 'start': 17401.323, 'duration': 1.661}, {'end': 17407.988, 'text': 'Now, can you just put 1 million images in the forward and neural network? No, never.', 'start': 17403.385, 'duration': 4.603}, {'end': 17412.511, 'text': 'You have to give them like 1, 000, 10, 000 images, one at a time, right?', 'start': 17408.849, 'duration': 3.662}, {'end': 17413.932, 'text': 'Because our RAMs are limited, right?', 'start': 17412.551, 'duration': 1.381}, {'end': 17417.535, 'text': "We still don't have processing power that can take all the images.", 'start': 17414.813, 'duration': 2.722}], 'summary': 'Using batches to process 1 million images due to limited ram and processing power.', 'duration': 25.693, 'max_score': 17391.842, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/W7N6LPp0SmY/pics/W7N6LPp0SmY17391842.jpg'}, {'end': 17528.534, 'src': 'embed', 'start': 17497.779, 'weight': 3, 'content': [{'end': 17499.96, 'text': "Are you getting my point what I'm trying to say? This is pure theory.", 'start': 17497.779, 'duration': 2.181}, {'end': 17505.658, 'text': "that if you're giving a very less number of batches, then it's difficult for the model to run.", 'start': 17500.814, 'duration': 4.844}, {'end': 17514.544, 'text': 'But if you are giving a very large number, then you are forcing the model to learn multiple features in just one.', 'start': 17506.158, 'duration': 8.386}, {'end': 17515.365, 'text': 'go right?', 'start': 17514.544, 'duration': 0.821}, {'end': 17519.087, 'text': "That's why you have to decide the number very precisely.", 'start': 17516.225, 'duration': 2.862}, {'end': 17524.551, 'text': 'based on how many features in your images do you have, how difficult your image is to categorize?', 'start': 17519.087, 'duration': 5.464}, {'end': 17528.534, 'text': "or let's say that if you're training on two images, is the difference very significant?", 'start': 17524.551, 'duration': 3.983}], 'summary': 'Balancing batch size is crucial for model training; consider image complexity and number of features.', 'duration': 30.755, 'max_score': 17497.779, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/W7N6LPp0SmY/pics/W7N6LPp0SmY17497779.jpg'}, {'end': 17670.485, 'src': 'embed', 'start': 17635.445, 'weight': 7, 'content': [{'end': 17637.267, 'text': 'Yeah, actually, I am not saying about resizing.', 'start': 17635.445, 'duration': 1.822}, {'end': 17647.735, 'text': 'Resizing is also one of the factor, because if I resize the image to very small size means suppose 1000 by 1000 pixel image,', 'start': 17637.747, 'duration': 9.988}, {'end': 17658.164, 'text': 'if I reduce it to 20 by 20 pixel image, then the processing speed for that entire data will be fast enough right?', 'start': 17647.735, 'duration': 10.429}, {'end': 17661.407, 'text': 'But it will minimize the features in that.', 'start': 17658.625, 'duration': 2.782}, {'end': 17670.485, 'text': 'so if i reduce to that to some 500 pixel, to 500 pixel, then, uh, the uh, it will take some more processing time.', 'start': 17662.438, 'duration': 8.047}], 'summary': 'Resizing images can affect processing speed and feature preservation.', 'duration': 35.04, 'max_score': 17635.445, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/W7N6LPp0SmY/pics/W7N6LPp0SmY17635445.jpg'}, {'end': 17867.205, 'src': 'embed', 'start': 17840.16, 'weight': 5, 'content': [{'end': 17846.681, 'text': "And if you're working on wires, they're very minute and you know it's even difficult for humans to differentiate.", 'start': 17840.16, 'duration': 6.521}, {'end': 17849.102, 'text': 'and how can the model differentiate right?', 'start': 17846.681, 'duration': 2.421}, {'end': 17850.262, 'text': 'So the first thing is data.', 'start': 17849.122, 'duration': 1.14}, {'end': 17853.963, 'text': 'Your data is the first thing on which you can decide this number.', 'start': 17851.282, 'duration': 2.681}, {'end': 17860.904, 'text': 'Second is that people have made generic models like VGGNet, ResNet, and ImageNet.', 'start': 17854.283, 'duration': 6.621}, {'end': 17867.205, 'text': 'So what I generally do is that I take inspiration from these papers, from these models right?', 'start': 17861.264, 'duration': 5.941}], 'summary': 'Data is crucial for model differentiation, leveraging generic models like vggnet and resnet for inspiration.', 'duration': 27.045, 'max_score': 17840.16, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/W7N6LPp0SmY/pics/W7N6LPp0SmY17840160.jpg'}, {'end': 17952.548, 'src': 'embed', 'start': 17922.492, 'weight': 4, 'content': [{'end': 17926.654, 'text': 'Right? Then you will see a graph somewhat like this at this section.', 'start': 17922.492, 'duration': 4.162}, {'end': 17929.675, 'text': 'It is almost like an elbow.', 'start': 17928.354, 'duration': 1.321}, {'end': 17938.921, 'text': 'our hand elbow somewhat like that, and this is your optimized batch size that you should take.', 'start': 17930.396, 'duration': 8.525}, {'end': 17944.364, 'text': 'this is the optimized batch size that you should always take.', 'start': 17938.921, 'duration': 5.443}, {'end': 17945.704, 'text': 'so first thing is your data.', 'start': 17944.364, 'duration': 1.34}, {'end': 17949.066, 'text': 'your data will tell you that how much you have to resize it.', 'start': 17945.704, 'duration': 3.362}, {'end': 17952.548, 'text': "so, uh, i'll tell you how do i decided that i?", 'start': 17949.066, 'duration': 3.482}], 'summary': 'Optimize batch size based on data for improved performance.', 'duration': 30.056, 'max_score': 17922.492, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/W7N6LPp0SmY/pics/W7N6LPp0SmY17922492.jpg'}], 'start': 17059.305, 'title': 'Tensorflow and image processing in neural networks', 'summary': 'Covers tensorflow session initialization, batch processing, and workflow, emphasizing 550 iterations for 55,000 training images. it also discusses the significance of batch size, image resizing, and model optimization using generic models like vggnet and resnet and the elbow algorithm.', 'chapters': [{'end': 17332.44, 'start': 17059.305, 'title': 'Tensorflow session and batch processing', 'summary': 'Discusses the initialization of global variables, running a training session with batch processing, and understanding the workflow of a session in tensorflow, emphasizing the need for 550 iterations to process 55,000 training images in 100-image batches.', 'duration': 273.135, 'highlights': ['Training with 55,000 images in 100-image batches requires 550 iterations. By processing 55,000 images in 100-image batches, it necessitates 550 iterations for complete processing.', 'Understanding the workflow of a session in TensorFlow. Explaining the sequence of operations within a TensorFlow session, involving placeholders, variable initialization, and the execution of a training step.', 'Initialization of global variables before running a session. Emphasizing the importance of initializing global variables as the first step before executing a session in TensorFlow.']}, {'end': 17777.641, 'start': 17332.992, 'title': 'Batch size & image resizing in neural networks', 'summary': 'Explains the importance of batch size in neural networks due to limited ram resources, and how resizing images affects processing speed and feature retention, emphasizing the need for a precise ratio of batch size to features.', 'duration': 444.649, 'highlights': ["The concept of batches came because we don't have as much as RAM we can afford. Explains the reason for using batches due to limited RAM resources.", 'Resizing images to smaller sizes can increase processing speed but may minimize features. Highlights the impact of resizing images on processing speed and feature retention.', 'Emphasizes the need for a precise ratio of batch size to features for generalized model training. Stresses the importance of a balanced batch size to ensure generalized model training.']}, {'end': 18245.416, 'start': 17777.641, 'title': 'Optimizing image recognition models', 'summary': "Discusses the importance of data in determining the model's performance, the use of generic models like vggnet and resnet for inspiration, and the use of the elbow algorithm to determine the optimized batch size and image dimensions for model training.", 'duration': 467.775, 'highlights': ["The importance of data in determining the model's performance The chapter emphasizes the impact of data on model performance, highlighting the differentiation between nut and bolt classification and minute wiring classification.", 'Use of generic models like VGGNet and ResNet for inspiration The speaker mentions drawing inspiration from established models like VGGNet and ResNet, demonstrating a practical approach to model development.', 'Utilization of the elbow algorithm to determine optimized batch size and image dimensions The chapter explains the application of the elbow algorithm to find the optimized batch size and image dimensions, offering a method for efficient model training.']}], 'duration': 1186.111, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/W7N6LPp0SmY/pics/W7N6LPp0SmY17059305.jpg', 'highlights': ['Training with 55,000 images in 100-image batches requires 550 iterations.', 'Understanding the workflow of a session in TensorFlow.', 'Initialization of global variables before running a session.', 'Emphasizes the need for a precise ratio of batch size to features for generalized model training.', 'Utilization of the elbow algorithm to determine optimized batch size and image dimensions.', 'Use of generic models like VGGNet and ResNet for inspiration.', "The importance of data in determining the model's performance.", 'Resizing images to smaller sizes can increase processing speed but may minimize features.', "The concept of batches came because we don't have as much as RAM we can afford."]}, {'end': 19726.271, 'segs': [{'end': 18273.848, 'src': 'embed', 'start': 18245.876, 'weight': 2, 'content': [{'end': 18248.597, 'text': 'So what I had to do, I had to crop this section first of all.', 'start': 18245.876, 'duration': 2.721}, {'end': 18253.973, 'text': 'The small section, I had to crop the small section and then I had to give it to model.', 'start': 18249.369, 'duration': 4.604}, {'end': 18256.795, 'text': 'And then only I could do it.', 'start': 18255.774, 'duration': 1.021}, {'end': 18261.619, 'text': 'Right? Yeah.', 'start': 18259.637, 'duration': 1.982}, {'end': 18264.961, 'text': 'So your question is totally valid and I can totally relate to it.', 'start': 18261.939, 'duration': 3.022}, {'end': 18266.763, 'text': 'I had the same questions.', 'start': 18265.141, 'duration': 1.622}, {'end': 18271.286, 'text': 'I did the same questions and I also could not get it.', 'start': 18267.003, 'duration': 4.283}, {'end': 18273.848, 'text': 'But believe me, but believe me, slowly, slowly, slowly.', 'start': 18271.326, 'duration': 2.522}], 'summary': 'The speaker had to crop a small section and give it to a model to accomplish a task, and acknowledges the challenges faced.', 'duration': 27.972, 'max_score': 18245.876, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/W7N6LPp0SmY/pics/W7N6LPp0SmY18245876.jpg'}, {'end': 18419.625, 'src': 'embed', 'start': 18392.696, 'weight': 0, 'content': [{'end': 18398.48, 'text': 'So you should definitely go with these things because they are trained to be very generic in nature.', 'start': 18392.696, 'duration': 5.784}, {'end': 18403.723, 'text': 'So they have almost 1000 categories to classify and they are trained on 1 million images.', 'start': 18399.04, 'duration': 4.683}, {'end': 18409.677, 'text': 'So you should definitely take inspiration from these things.', 'start': 18405.333, 'duration': 4.344}, {'end': 18416.362, 'text': 'I generally take the inspiration from these things because these are made by one of the greatest people in deep learning and AI section.', 'start': 18410.497, 'duration': 5.865}, {'end': 18419.625, 'text': 'And the second part is that they are very generic in nature.', 'start': 18416.963, 'duration': 2.662}], 'summary': 'The models are trained on 1 million images and have almost 1000 categories for classification, providing a strong foundation for inspiration.', 'duration': 26.929, 'max_score': 18392.696, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/W7N6LPp0SmY/pics/W7N6LPp0SmY18392696.jpg'}, {'end': 18759.813, 'src': 'embed', 'start': 18733.598, 'weight': 3, 'content': [{'end': 18740.862, 'text': 'close, very important, just do this right.', 'start': 18733.598, 'duration': 7.264}, {'end': 18744.564, 'text': 'so here see, my intention today was not to make a model out of it.', 'start': 18740.862, 'duration': 3.702}, {'end': 18751.168, 'text': "it was just to show you that how, uh, how low level writing you have to do while you're writing in your.", 'start': 18744.564, 'duration': 6.604}, {'end': 18754.39, 'text': "uh, while you're writing in your tensorflow code.", 'start': 18751.168, 'duration': 3.222}, {'end': 18759.813, 'text': 'right, this is the same code that we wrote last in the last class, in just, i mean 10 or 20 minutes.', 'start': 18754.39, 'duration': 5.423}], 'summary': 'Demonstrated low-level writing in tensorflow code in 10-20 minutes.', 'duration': 26.215, 'max_score': 18733.598, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/W7N6LPp0SmY/pics/W7N6LPp0SmY18733598.jpg'}, {'end': 19051.843, 'src': 'embed', 'start': 19012.625, 'weight': 4, 'content': [{'end': 19025.839, 'text': 'yeah, see, as you told, like that, in reshaping the image the features are lost.', 'start': 19012.625, 'duration': 13.214}, {'end': 19031.686, 'text': 'so when we reshape so that will impact the output probability also, right.', 'start': 19025.839, 'duration': 5.847}, {'end': 19045.477, 'text': 'so we have to uh, for uh, you know performance and do we have to do iterations like how much reshape and what is the output coming and change?', 'start': 19032.685, 'duration': 12.792}, {'end': 19051.843, 'text': 'keep on changing the reshape thing, see again.', 'start': 19045.477, 'duration': 6.366}], 'summary': 'Reshaping image impacts output probability, requiring performance iterations.', 'duration': 39.218, 'max_score': 19012.625, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/W7N6LPp0SmY/pics/W7N6LPp0SmY19012625.jpg'}, {'end': 19428.237, 'src': 'embed', 'start': 19384.491, 'weight': 1, 'content': [{'end': 19392.577, 'text': "there are two concepts transfer learning and fine-tuning in image classifier in image, in computer vision, and I'll tell you that also, don't worry,", 'start': 19384.491, 'duration': 8.086}, {'end': 19408.228, 'text': "don't worry, that is in our, in our course, make sense, yes, cool.", 'start': 19392.577, 'duration': 15.651}, {'end': 19409.909, 'text': 'so now, now what we will do.', 'start': 19408.228, 'duration': 1.681}, {'end': 19422.649, 'text': "is we go to this cool website here and I'll just copy image address.", 'start': 19409.909, 'duration': 12.74}, {'end': 19424.452, 'text': 'okay, this is benchmark only.', 'start': 19422.649, 'duration': 1.803}, {'end': 19428.237, 'text': 'okay, ctrl shift t.', 'start': 19424.452, 'duration': 3.785}], 'summary': 'Transcript covers transfer learning and fine-tuning in image classifier in computer vision.', 'duration': 43.746, 'max_score': 19384.491, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/W7N6LPp0SmY/pics/W7N6LPp0SmY19384491.jpg'}], 'start': 18245.876, 'title': 'Ai implementation challenges and techniques', 'summary': 'Discusses challenges and use cases in ai implementation, emphasizing the need for cropping sections, the value of generic models like vgg, resnet, and inception in deep learning, the usage of tensorflow and keras for building a perceptron model, and the impact of reshaping images on model performance.', 'chapters': [{'end': 18294.633, 'start': 18245.876, 'title': 'Challenges in ai implementation', 'summary': 'Discusses the challenges and interesting use cases in ai implementation, highlighting the need for cropping sections and the curiosity around solving problems with ai.', 'duration': 48.757, 'highlights': ['The chapter discusses the need to crop small sections for AI modeling and the challenges faced in implementing AI (3 instances).', 'It emphasizes the interesting use cases in AI and the curiosity around solving problems with AI (2 instances).', 'The speaker shares personal experiences and relates to the challenges faced in implementing AI (1 instance).']}, {'end': 18630.221, 'start': 18294.633, 'title': 'Choosing generic models for deep learning', 'summary': 'Discusses the importance of using generic models like vgg, resnet, and inception in deep learning, which are trained on 1 million images and have almost 1000 categories to classify, emphasizing their value for generalization and inspiration in ai.', 'duration': 335.588, 'highlights': ['The importance of using generic models like VGG, ResNet, and Inception in deep learning, which are trained on 1 million images and have almost 1000 categories to classify.', 'Emphasizing the value of these models for generalization and inspiration in AI.', 'The limitations of seeking correct answers in multiple blogs and books, as they are often tailored to specific problems and may not provide a universal solution.', 'Encouraging reliance on models such as VGG, ResNet, and Inception, created by leading experts in deep learning and AI, for their generic nature and wide applicability.', 'Discussion about training a model using a batch of 100 images and the process of assessing accuracy through TF dot equals and argmax functions.']}, {'end': 19012.625, 'start': 18630.221, 'title': 'Introduction to tensorflow and keras', 'summary': 'Introduces the usage of tensorflow and keras for building a perceptron model with an accuracy of 0.09, while emphasizing the importance of low-level writing in tensorflow and expressing a preference for keras due to its ease of use and control.', 'duration': 382.404, 'highlights': ['The chapter emphasizes the importance of low-level writing in TensorFlow and expresses a preference for Keras due to its ease of use and control.', 'A perceptron model with an accuracy of 0.09 is built using TensorFlow, showcasing the functionality of the platform.', 'The instructor discusses the challenges of working with images and expresses a preference for covering NLP examples due to the complexities of pre-processing and post-processing in NLP.', 'The instructor acknowledges the difficulty in working with images and suggests exploring examples related to NLP and chatbots for a better understanding of deep learning applications.']}, {'end': 19726.271, 'start': 19012.625, 'title': 'Optimizing image reshaping for model performance', 'summary': 'Discusses the impact of reshaping images on output probability, suggests using elbo requirements to determine optimized batch size, and emphasizes the use of transfer learning and fine-tuning for image classification.', 'duration': 713.646, 'highlights': ['The impact of reshaping images on output probability Reshaping images can impact the output probability, requiring iterations to determine the optimal reshaping for model performance.', 'Using ELBO requirements to determine optimized batch size ELBO requirements involve plotting error against batch size to find the optimized batch size for the computer architecture, based on minimum error.', 'Emphasizing transfer learning and fine-tuning for image classification The use of pre-trained models and transfer learning is recommended for image classification, along with fine-tuning to leverage existing learning from pre-trained models.']}], 'duration': 1480.395, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/W7N6LPp0SmY/pics/W7N6LPp0SmY18245876.jpg', 'highlights': ['The importance of using generic models like VGG, ResNet, and Inception in deep learning, which are trained on 1 million images and have almost 1000 categories to classify.', 'Emphasizing transfer learning and fine-tuning for image classification The use of pre-trained models and transfer learning is recommended for image classification, along with fine-tuning to leverage existing learning from pre-trained models.', 'The chapter discusses the need to crop small sections for AI modeling and the challenges faced in implementing AI (3 instances).', 'The chapter emphasizes the importance of low-level writing in TensorFlow and expresses a preference for Keras due to its ease of use and control.', 'The impact of reshaping images on output probability Reshaping images can impact the output probability, requiring iterations to determine the optimal reshaping for model performance.']}, {'end': 21249.294, 'segs': [{'end': 19829.096, 'src': 'embed', 'start': 19795.606, 'weight': 0, 'content': [{'end': 19803.531, 'text': 'Keras is using this data set to load the images which it already had.', 'start': 19795.606, 'duration': 7.925}, {'end': 19811.957, 'text': 'Like in some company, they want to create their own data set and own images for whatever data.', 'start': 19804.472, 'duration': 7.485}, {'end': 19815.992, 'text': 'So how do you do that? Very good.', 'start': 19813.291, 'duration': 2.701}, {'end': 19817.892, 'text': 'I totally understand.', 'start': 19817.072, 'duration': 0.82}, {'end': 19820.093, 'text': 'And I know, I know, I know.', 'start': 19818.433, 'duration': 1.66}, {'end': 19824.494, 'text': 'This is just starting, right? And today, we are going to do this.', 'start': 19820.113, 'duration': 4.381}, {'end': 19829.096, 'text': 'And tomorrow, the session that I have planned, we have only images.', 'start': 19824.514, 'duration': 4.582}], 'summary': "Keras uses existing dataset for image loading. tomorrow's session focuses on image processing.", 'duration': 33.49, 'max_score': 19795.606, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/W7N6LPp0SmY/pics/W7N6LPp0SmY19795606.jpg'}, {'end': 20149.579, 'src': 'embed', 'start': 20121.867, 'weight': 3, 'content': [{'end': 20125.228, 'text': 'So to avoid these things what you do you normalize your image.', 'start': 20121.867, 'duration': 3.361}, {'end': 20128.75, 'text': "Now I'll tell you what does the normalize means.", 'start': 20126.709, 'duration': 2.041}, {'end': 20137.433, 'text': 'Normalize if you are asking in mathematics normalize means that you want to squeeze you want to squeeze your data into a particular boundary.', 'start': 20129.27, 'duration': 8.163}, {'end': 20142.875, 'text': 'Right, you want to let it follow some kind of a normal curve.', 'start': 20138.113, 'duration': 4.762}, {'end': 20149.579, 'text': 'so if you have done, if you know a little bit stats about this thing, this is this is called a bell-shaped curve.', 'start': 20142.875, 'duration': 6.704}], 'summary': 'Normalize image data to fit within a specific boundary, following a bell-shaped curve.', 'duration': 27.712, 'max_score': 20121.867, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/W7N6LPp0SmY/pics/W7N6LPp0SmY20121867.jpg'}, {'end': 20467.696, 'src': 'embed', 'start': 20443.63, 'weight': 2, 'content': [{'end': 20451.977, 'text': 'dense layer is nothing, but that is that it makes sure that you are connecting every previous layer to the to the next layer,', 'start': 20443.63, 'duration': 8.347}, {'end': 20459.27, 'text': 'so that is also called as a fully connected layer right Next, what I am going to do is I am going to use flatten layer.', 'start': 20451.977, 'duration': 7.293}, {'end': 20461.712, 'text': "I will tell you don't worry what is a flatten layer.", 'start': 20459.29, 'duration': 2.422}, {'end': 20465.554, 'text': 'Then we have keras dot optimizers.', 'start': 20462.692, 'duration': 2.862}, {'end': 20467.696, 'text': 'I will use SGD.', 'start': 20466.075, 'duration': 1.621}], 'summary': 'The transcript discusses connecting layers in a neural network and using the sgd optimizer in keras.', 'duration': 24.066, 'max_score': 20443.63, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/W7N6LPp0SmY/pics/W7N6LPp0SmY20443630.jpg'}, {'end': 20974.443, 'src': 'embed', 'start': 20944.454, 'weight': 1, 'content': [{'end': 20948.356, 'text': 'So one image will go Sanket was asking this question that how does this batch size.', 'start': 20944.454, 'duration': 3.902}, {'end': 20952.918, 'text': 'So batch size one means that only one image is going forward and backward.', 'start': 20948.656, 'duration': 4.262}, {'end': 20957.36, 'text': 'Now imagine Sanket how difficult it is for the model to understand with just one image right.', 'start': 20953.038, 'duration': 4.322}, {'end': 20959.653, 'text': 'How it is going to update the weights.', 'start': 20957.952, 'duration': 1.701}, {'end': 20964.897, 'text': 'So if you understand the theory about deep neural networks.', 'start': 20960.013, 'duration': 4.884}, {'end': 20967.959, 'text': 'Just close your eyes and go inside the deep neural networks.', 'start': 20964.937, 'duration': 3.022}, {'end': 20970.14, 'text': 'And now imagine that one image is coming.', 'start': 20968.359, 'duration': 1.781}, {'end': 20974.443, 'text': 'Now one image is coming and you are trying to learn about this image.', 'start': 20970.961, 'duration': 3.482}], 'summary': 'Training with a batch size of one image makes it difficult for the model to update weights effectively in deep neural networks.', 'duration': 29.989, 'max_score': 20944.454, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/W7N6LPp0SmY/pics/W7N6LPp0SmY20944454.jpg'}], 'start': 19726.451, 'title': 'Image data processing and model building', 'summary': 'Discusses tuple unfolding, data normalization, sequential model building with keras, and batch size understanding. it covers examples of 60,000 images for classification and achieving 85% training and 83% testing accuracy with an error rate of 0.45%.', 'chapters': [{'end': 19957.812, 'start': 19726.451, 'title': 'Unfolding of tuple in data set', 'summary': 'Discusses unpacking tuples, loading and classifying images in data sets, with examples of 60,000 images and plans for traffic signal and animal image classification.', 'duration': 231.361, 'highlights': ['The chapter covers unpacking of tuples and loading of images in a data set, including examples of 60,000 images and corresponding labels.', 'There are plans for classification of animal images and traffic signal examples, showcasing different structures and challenges in classifying images.', 'The instructor reassures about having data sets for animal images and traffic signal examples, emphasizing preparedness and addressing potential challenges in classifying images.', 'The discussion includes unpacking of tuples and loading of images in a data set, with examples of 60,000 images and corresponding labels.']}, {'end': 20414.201, 'start': 19957.812, 'title': 'Image data normalization', 'summary': 'Explains the importance of normalizing image data to avoid exploding gradients, using the min-max scalar technique to squeeze pixel values from 0 to 255 into the range of 0 to 1, ensuring that the data remains the same.', 'duration': 456.389, 'highlights': ['Explaining the need for data normalization to avoid exploding gradients and biased model towards larger values. The chapter emphasizes the importance of normalizing image data to avoid exploding gradients and biased models towards larger values, highlighting the need for data normalization to prevent these issues.', 'Utilizing the min-max scalar technique to squeeze pixel values from 0 to 255 into the range of 0 to 1. The chapter explains the min-max scalar technique, where pixel values from 0 to 255 are divided by the maximum value to squeeze them into the range of 0 to 1, ensuring that the data integrity and ratio remain the same.', 'Demonstrating the impact of normalization on image data by comparing the normalized and original images. The chapter demonstrates the impact of normalization on image data by comparing the normalized and original images, showing that the ratio between the two images remains the same after normalization.']}, {'end': 20922.24, 'start': 20414.641, 'title': 'Building sequential model with keras', 'summary': 'Covers the process of building a sequential model with keras, including the use of dense layers, the concept of flattening images, and the application of sgd optimizer and categorical cross entropy loss, with the aim of achieving accuracy through model fitting for 10 epochs.', 'duration': 507.599, 'highlights': ['Building a sequential model with Keras The chapter focuses on the process of building a sequential model with Keras, leveraging dense layers and flattening images to accommodate different image dimensions, and using SGD optimizer and categorical cross entropy loss for model compilation.', "Application of SGD optimizer and categorical cross entropy loss The usage of SGD optimizer and categorical cross entropy loss is emphasized for model compilation, aimed at optimizing the model's performance and accuracy.", 'Process of model fitting for 10 epochs The chapter details the process of model fitting for 10 epochs, indicating the iterative nature of training the model to improve its accuracy and performance.', 'Concept of flattening images for different image dimensions The concept of flattening images is explained to adapt to different image dimensions, such as converting 28x28 images into a format suitable for the model to process.', 'Explanation of ReLU activation function The chapter provides an explanation of the ReLU activation function, highlighting its role in handling negative values and its recommendation for usage in model building.']}, {'end': 21249.294, 'start': 20923.381, 'title': 'Understanding batch size in neural networks', 'summary': "Explains the importance of batch size in neural networks, emphasizing how it affects the model's training by updating weights based on a batch of images, and demonstrates the impact of batch size on training and testing accuracy, achieving 85% accuracy on training and 83% on testing with an error rate of 0.45%.", 'duration': 325.913, 'highlights': ['The importance of batch size in neural networks is emphasized, illustrating how a smaller batch size can lead to biased weight updates based on individual images, while a larger batch size provides a more random cluster of images for better learning, with the demonstration of increasing batch size leading to improved modeling and accuracy on testing reaching 83%.', 'The impact of batch size on training and testing accuracy is demonstrated, achieving 85% accuracy on training and 83% on testing with an error rate of 0.45%, showcasing the significance of batch size in influencing model performance.', 'The explanation of batch size in neural networks is provided, illustrating how it determines the number of images processed in both forward and backward propagation, with the example of 10 images in a batch leading to 10 times of going through 100 images, highlighting the practical application of batch size in model training.']}], 'duration': 1522.843, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/W7N6LPp0SmY/pics/W7N6LPp0SmY19726451.jpg', 'highlights': ['The chapter covers unpacking of tuples and loading of images in a data set, including examples of 60,000 images and corresponding labels.', 'The importance of batch size in neural networks is emphasized, illustrating how a smaller batch size can lead to biased weight updates based on individual images, while a larger batch size provides a more random cluster of images for better learning, with the demonstration of increasing batch size leading to improved modeling and accuracy on testing reaching 83%.', 'Building a sequential model with Keras The chapter focuses on the process of building a sequential model with Keras, leveraging dense layers and flattening images to accommodate different image dimensions, and using SGD optimizer and categorical cross entropy loss for model compilation.', 'Explaining the need for data normalization to avoid exploding gradients and biased model towards larger values. The chapter emphasizes the importance of normalizing image data to avoid exploding gradients and biased models towards larger values, highlighting the need for data normalization to prevent these issues.']}], 'highlights': ["AI's wide applications in medical science have led to the creation of virtual personal healthcare assistants and efficient healthcare bots, revolutionizing healthcare support for patients.", 'The tutorial covers ai basics, impact across industries, machine learning, deep learning, tensorflow basics, neural networks, image processing, mnist, multi-class image classification, and ai implementation challenges.', 'The session covers major concepts of artificial intelligence, including AI, machine learning, and deep learning.', 'The agenda includes understanding the topology of a neural network and training the network with backpropagation.', 'The session concludes with implementation demos on the MNIST dataset with TensorFlow and Fashion dataset with Keras.', "AI's applications in banking and finance enable efficient results and quick resolution, reducing time and effort from employees.", 'Artificial intelligence enables machines to exhibit human-like intelligence, including the abilities to think, learn, and make decisions, which is truly amazing.', "AI's applications in aerospace make air transport efficient, fast, safe, and provide a comfortable journey to passengers.", "AI's definition and existing applications, including chatbots like OK Google and Siri, and humanoid robots like Sophia, along with self-driving cars by Google and Tesla, demonstrate its wide-ranging impact and potential.", 'Unsupervised learning involves model learning through observation and finding structures in the data set by creating clusters in it.', 'The limitations of machine learning are discussed, including the need for massive training data, difficulty in error diagnosis, lack of creativity, and time constraints.', 'Deep learning is introduced as a subset of machine learning that learns through data representations and utilizes deep neural networks.', 'Tensors are the building blocks in TensorFlow, representing data in the form of multi-dimensional arrays.', 'The rank of a tensor determines the number of dimensions used to represent the data.', "Duffy Hilton's success in the ImageNet competition in 2004 with a dataset of 1 million images covering 1000 categories.", 'The breakthrough in neural network accuracy reached 94% in 2004, surpassing human accuracy.', "The concept of backpropagation involves adjusting the weights of the neural network based on the error between the actual output and the predicted output, aiming to minimize the error and improve the model's accuracy.", "Determining the optimal number of layers and neurons is crucial, involving hit-and-try and understanding the model's capacity for efficient performance.", 'Backpropagation is crucial for AI and serves as the main backbone for learning from errors.', 'The iterative adjustment of w1 and w2 values minimizes error, leading to the error graph saturating and becoming parallel to the x-axis.', 'Major frameworks like Keras and TensorFlow default to a learning rate value of 0.001, underlining its significance in model training.', 'The concept of image pixel values, ranging from 0 to 255, is explained, emphasizing the importance of understanding these values in image processing.', 'The MNIST dataset consists of 55,000 training images, 5,000 validation images, and 10,000 testing images, with each image containing 784 pixels.', 'The softmax activation function aids in decision-making for multi-class classification tasks.', 'The process of reshaping the flattened image data into 28x28 dimensions is explained, aligning with the standard image size in the MNIST dataset.', 'The importance of using generic models like VGG, ResNet, and Inception in deep learning, which are trained on 1 million images and have almost 1000 categories to classify.', 'The chapter covers unpacking of tuples and loading of images in a data set, including examples of 60,000 images and corresponding labels.', 'The importance of batch size in neural networks is emphasized, illustrating how a smaller batch size can lead to biased weight updates based on individual images, while a larger batch size provides a more random cluster of images for better learning, with the demonstration of increasing batch size leading to improved modeling and accuracy on testing reaching 83%.']}