title
Lesson 1: Practical Deep Learning for Coders 2022

description
Go to https://course.fast.ai for code, notebooks, quizzes, etc. This course is designed for people with some coding experience who want to learn how to apply deep learning and machine learning to practical problems. There are 9 lessons, and each lesson is around 90 minutes long. We cover topics such as how to: - Build and train deep learning, random forest, and regression models - Deploy models - Apply deep learning to computer vision, natural language processing, tabular analysis, and collaborative filtering problems - Use PyTorch, the world’s fastest growing deep learning software, together with popular libraries such as fastai, Hugging Face Transformers, and gradio You don’t need any special hardware or software — we’ll show you how to use free resources for both building and deploying models. You don’t need any university math either — we’ll teach you the calculus and linear algebra you need during the course. 00:00 - Introduction 00:25 - What has changed since 2015 01:20 - Is it a bird 02:09 - Images are made of numbers 03:29 - Downloading images 04:25 - Creating a DataBlock and Learner 05:18 - Training the model and making a prediction 07:20 - What can deep learning do now 10:33 - Pathways Language Model (PaLM) 15:40 - How the course will be taught. Top down learning 19:25 - Jeremy Howard’s qualifications 22:38 - Comparison between modern deep learning and 2012 machine learning practices 24:31 - Visualizing layers of a trained neural network 27:40 - Image classification applied to audio 28:08 - Image classification applied to time series and fraud 30:16 - Pytorch vs Tensorflow 31:43 - Example of how Fastai builds off Pytorch (AdamW optimizer) 35:18 - Using cloud servers to run your notebooks (Kaggle) 38:45 - Bird or not bird? & explaining some Kaggle features 40:15 - How to import libraries like Fastai in Python 40:42 - Best practice - viewing your data between steps 42:00 - Datablocks API overarching explanation 44:40 - Datablocks API parameters explanation 48:40 - Where to find fastai documentation 49:54 - Fastai’s learner (combines model & data) 50:40 - Fastai’s available pretrained models 52:02 - What’s a pretrained model? 53:48 - Testing your model with predict method 55:08 - Other applications of computer vision. Segmentation 56:48 - Segmentation code explanation 58:32 - Tabular analysis with fastai 59:42 - show_batch method explanation 1:01:25 - Collaborative filtering (recommendation system) example 1:05:08 - How to turn your notebooks into a presentation tool (RISE) 1:05:45 - What else can you make with notebooks? 1:08:06 - What can deep learning do presently? 1:10:33 - The first neural network - Mark I Perceptron (1957) 1:12:38 - Machine learning models at a high level 1:18:27 - Homework Thanks to bencoman, mike.moloch, amr.malik, and gagan on forums.fast.ai for creating the transcript. Thanks to Raymond-Wu on forums.fast.ai for help with chapter titles.

detail
{'title': 'Lesson 1: Practical Deep Learning for Coders 2022', 'heatmap': [{'end': 2392.195, 'start': 2285.753, 'weight': 0.872}, {'end': 2839.914, 'start': 2684.578, 'weight': 0.807}, {'end': 3587.214, 'start': 3529.065, 'weight': 0.71}], 'summary': "Practical deep learning lesson 1 explores image recognition, creating a computer vision model using fastai to classify 200 pictures of birds and forests in under 30 seconds. the course delves into deep learning applications in art, language models, python as a powerful calculator, pytorch's dominance over tensorflow, data preparation, pytorch data loaders, and deep learning applications in various fields.", 'chapters': [{'end': 597.008, 'segs': [{'end': 33.057, 'src': 'embed', 'start': 2.096, 'weight': 2, 'content': [{'end': 6.041, 'text': 'Welcome to practical deep learning for coders lesson one.', 'start': 2.096, 'duration': 3.945}, {'end': 16.033, 'text': "This is version five of this course and it's the first new one we've done in two years, so we've got a lot of cool things to cover.", 'start': 6.481, 'duration': 9.552}, {'end': 19.477, 'text': "It's amazing how much has changed.", 'start': 17.855, 'duration': 1.622}, {'end': 25.833, 'text': 'here is a XKCD from the end of 2015.', 'start': 22.031, 'duration': 3.802}, {'end': 33.057, 'text': 'Who here has seen XKCD comics before? Pretty much everybody, not surprising.', 'start': 25.833, 'duration': 7.224}], 'summary': 'Practical deep learning for coders, version five, covers a lot of cool things in the first new course in two years.', 'duration': 30.961, 'max_score': 2.096, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/8SF_h3xF3cE/pics/8SF_h3xF3cE2096.jpg'}, {'end': 100.68, 'src': 'embed', 'start': 68.609, 'weight': 4, 'content': [{'end': 72.832, 'text': "We're now going to build exactly that system for free in about two minutes.", 'start': 68.609, 'duration': 4.223}, {'end': 77.056, 'text': "So let's build an is-it-a-bird system.", 'start': 75.454, 'duration': 1.602}, {'end': 83.968, 'text': "So we're going to use Python and so I'm going to run through this really quickly.", 'start': 79.645, 'duration': 4.323}, {'end': 91.133, 'text': "You're not expected to run through it with me because we're going to come back to it, okay, but let's go ahead and run that cell.", 'start': 83.988, 'duration': 7.145}, {'end': 100.68, 'text': "Okay, so what we're doing is we're searching DuckDuckGo for images of bird photos and we're just going to grab one.", 'start': 91.153, 'duration': 9.527}], 'summary': "Building a free 'is-it-a-bird' system using python in about two minutes.", 'duration': 32.071, 'max_score': 68.609, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/8SF_h3xF3cE/pics/8SF_h3xF3cE68609.jpg'}, {'end': 153.31, 'src': 'embed', 'start': 120.966, 'weight': 3, 'content': [{'end': 127.771, 'text': "Now we're going to need to build a system that can recognize things that are birds versus things that aren't birds from photos.", 'start': 120.966, 'duration': 6.805}, {'end': 136.418, 'text': 'Now of course computers need numbers to work with, but luckily images are made of numbers.', 'start': 128.312, 'duration': 8.106}, {'end': 153.31, 'text': "I actually found this really nice website called pixby, where I can grab a bird and if I wiggle over it, let's pick its beak.", 'start': 136.438, 'duration': 16.872}], 'summary': 'Developing a system to recognize birds in photos using pixby for image data.', 'duration': 32.344, 'max_score': 120.966, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/8SF_h3xF3cE/pics/8SF_h3xF3cE120966.jpg'}, {'end': 317.744, 'src': 'embed', 'start': 270.511, 'weight': 0, 'content': [{'end': 282.476, 'text': "so after I run this cell, you'll see that I've basically I'll go through the details of this later.", 'start': 270.511, 'duration': 11.965}, {'end': 289.495, 'text': 'but a data block gives fastai, the library, all the information it needs to create a computer vision model,', 'start': 282.476, 'duration': 7.019}, {'end': 296.539, 'text': "and so in this case we're basically telling it get all the image files that we just downloaded, and then we say show me a few, up to six,", 'start': 289.495, 'duration': 7.044}, {'end': 300.201, 'text': "and let's see, yeah, so we've got some birds, forest, bird, bird first.", 'start': 296.539, 'duration': 3.662}, {'end': 306.124, 'text': "okay, so one of the nice things about doing computer vision models is it's really easy to check your data because you can just look at it,", 'start': 300.201, 'duration': 5.923}, {'end': 308.606, 'text': 'which is not the case for a lot of kinds of models.', 'start': 306.124, 'duration': 2.482}, {'end': 317.744, 'text': "Okay, so we've now downloaded 200 pictures of birds, 200 pictures of forests.", 'start': 310.259, 'duration': 7.485}], 'summary': 'Using data block, fastai creates a computer vision model with 200 bird and 200 forest images.', 'duration': 47.233, 'max_score': 270.511, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/8SF_h3xF3cE/pics/8SF_h3xF3cE270511.jpg'}, {'end': 449.432, 'src': 'embed', 'start': 429.257, 'weight': 5, 'content': [{'end': 441.826, 'text': "So where have we got to now with deep learning? It moves so fast, but even in the last few weeks we've taken it up another notch as a community.", 'start': 429.257, 'duration': 12.569}, {'end': 449.432, 'text': 'You might have seen that something called DALI 2 has been released, which uses deep learning to generate new pictures.', 'start': 442.366, 'duration': 7.066}], 'summary': 'Deep learning community advances with release of dali 2, generating new pictures.', 'duration': 20.175, 'max_score': 429.257, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/8SF_h3xF3cE/pics/8SF_h3xF3cE429257.jpg'}], 'start': 2.096, 'title': 'Practical deep learning lesson 1 and creating computer vision models', 'summary': 'Introduces practical deep learning lesson 1, version five of the course, covering the transformation of image recognition and the process of obtaining and processing images. additionally, it discusses creating a computer vision model using the fastai library to classify 200 pictures of birds and 200 pictures of forests, running on a laptop in under 30 seconds.', 'chapters': [{'end': 246.223, 'start': 2.096, 'title': 'Practical deep learning lesson 1', 'summary': "Introduces the practical deep learning lesson 1, version five of the course, covering the transformation of image recognition, using python to build an 'is-it-a-bird' system in a few minutes, and the process of obtaining and processing images for the system.", 'duration': 244.127, 'highlights': ['The chapter introduces the practical deep learning lesson 1, version five of the course', "Using Python to build an 'is-it-a-bird' system in a few minutes", 'Process of obtaining and processing images for the system']}, {'end': 597.008, 'start': 246.784, 'title': 'Creating computer vision models', 'summary': 'Discusses the process of creating a computer vision model using fastai library to classify 200 pictures of birds and 200 pictures of forests, running on a laptop in under 30 seconds, demonstrating the accessibility and power of deep learning.', 'duration': 350.224, 'highlights': ['Creating a computer vision model to classify 200 pictures of birds and 200 pictures of forests, running on a laptop in under 30 seconds', 'Using fastai library to provide all information needed to create a computer vision model', 'Discussion about the rapid progress in deep learning, including the release of DALI 2 and its innovative applications']}], 'duration': 594.912, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/8SF_h3xF3cE/pics/8SF_h3xF3cE2096.jpg', 'highlights': ['Creating a computer vision model to classify 200 pictures of birds and 200 pictures of forests, running on a laptop in under 30 seconds', 'Using fastai library to provide all information needed to create a computer vision model', 'The chapter introduces the practical deep learning lesson 1, version five of the course', 'Process of obtaining and processing images for the system', "Using Python to build an 'is-it-a-bird' system in a few minutes", 'Discussion about the rapid progress in deep learning, including the release of DALI 2 and its innovative applications']}, {'end': 1081.816, 'segs': [{'end': 645.641, 'src': 'embed', 'start': 597.628, 'weight': 0, 'content': [{'end': 606.394, 'text': 'This guy is an artist who actually writes his own software based on deep learning and spends, you know, months on building stuff and, as you can see,', 'start': 597.628, 'duration': 8.766}, {'end': 607.675, 'text': 'you can really take it to the next level.', 'start': 606.394, 'duration': 1.281}, {'end': 618.462, 'text': "It's been really great actually to see how a lot of fast.ai alumni with backgrounds as artists have gone on to bring deep learning and art together,", 'start': 609.116, 'duration': 9.346}, {'end': 619.823, 'text': "and it's a very exciting direction.", 'start': 618.462, 'duration': 1.361}, {'end': 622.847, 'text': "it's not just images to be clear.", 'start': 621.266, 'duration': 1.581}, {'end': 630.692, 'text': "You know, one of another interesting thing that's popped up in the last couple of weeks is Google's pathways language model,", 'start': 622.867, 'duration': 7.825}, {'end': 643.419, 'text': 'which can take any arbitrary English as text question and can create an answer which not only answers the question but also explains its thinking,', 'start': 630.692, 'duration': 12.727}, {'end': 645.641, 'text': 'whatever it means for a language model to be thinking.', 'start': 643.419, 'duration': 2.222}], 'summary': "Artist writes own deep learning software, fast.ai alumni combining deep learning and art, google's pathways language model explained.", 'duration': 48.013, 'max_score': 597.628, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/8SF_h3xF3cE/pics/8SF_h3xF3cE597628.jpg'}, {'end': 720.515, 'src': 'embed', 'start': 688.305, 'weight': 2, 'content': [{'end': 693.507, 'text': 'if any of us would have believed would be maybe possible to do by computers even in our lifetime.', 'start': 688.305, 'duration': 5.202}, {'end': 702.925, 'text': 'This means that there is a lot of practical and ethical considerations.', 'start': 697.682, 'duration': 5.243}, {'end': 709.149, 'text': "We will touch on them during this course but can't possibly hope to do them justice.", 'start': 704.926, 'duration': 4.223}, {'end': 720.515, 'text': 'So I would certainly encourage you to check out ethics.fast.ai to see our whole data ethics course, taught by my co-founder, Dr.', 'start': 710.169, 'duration': 10.346}], 'summary': "Computers' capabilities exceed expectations, prompting ethical concerns. further exploration available at ethics.fast.ai", 'duration': 32.21, 'max_score': 688.305, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/8SF_h3xF3cE/pics/8SF_h3xF3cE688305.jpg'}, {'end': 774.075, 'src': 'embed', 'start': 748.652, 'weight': 3, 'content': [{'end': 753.776, 'text': 'and he has this great approach in his classrooms of figuring out how his students are getting along,', 'start': 748.652, 'duration': 5.124}, {'end': 763.483, 'text': "which is to put a colored cup on their desk green to mean that they're doing fine, yellow cup to mean I'm not quite sure,", 'start': 753.776, 'duration': 9.707}, {'end': 766.486, 'text': "and a red cup to mean I have no idea what's going on.", 'start': 763.483, 'duration': 3.003}, {'end': 774.075, 'text': "Now, since most of you are watching this remotely, I can't look at your caps and I don't think anybody bought colored caps with them today.", 'start': 767.026, 'duration': 7.049}], 'summary': 'Teacher uses colored cups to gauge student understanding in class.', 'duration': 25.423, 'max_score': 748.652, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/8SF_h3xF3cE/pics/8SF_h3xF3cE748652.jpg'}, {'end': 973.946, 'src': 'embed', 'start': 944.408, 'weight': 4, 'content': [{'end': 948.77, 'text': "We didn't start by doing an in-depth review of linear algebra and calculus.", 'start': 944.408, 'duration': 4.362}, {'end': 957.676, 'text': "That's because two of my favorite writers and researchers on education, Paul Lockhart and David Perkins,", 'start': 950.171, 'duration': 7.505}, {'end': 965, 'text': 'and many others talk about how much better people learn when they learn with a context in place.', 'start': 957.676, 'duration': 7.324}, {'end': 973.946, 'text': 'So the way we learn math at school, where we do like counting and then adding, and then fractions and then decimals, and then blah, blah, blah.', 'start': 965.501, 'duration': 8.445}], 'summary': 'Learning math with context enhances understanding and retention.', 'duration': 29.538, 'max_score': 944.408, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/8SF_h3xF3cE/pics/8SF_h3xF3cE944408.jpg'}, {'end': 1075.712, 'src': 'embed', 'start': 1043.31, 'weight': 5, 'content': [{'end': 1044.511, 'text': 'Just do your best to go along with it.', 'start': 1043.31, 'duration': 1.201}, {'end': 1047.352, 'text': "Those of you who haven't will find this very natural.", 'start': 1044.951, 'duration': 2.401}, {'end': 1053.014, 'text': 'Oh and this is Dylan Williams who I mentioned before, the guy who came up with the really cool cutscene.', 'start': 1048.353, 'duration': 4.661}, {'end': 1062.48, 'text': "There'll be a lot of tricks that have come out of the educational research literature scattered through this course.", 'start': 1054.574, 'duration': 7.906}, {'end': 1066.444, 'text': "On the whole I won't call them out, they'll just be there, but maybe from time to time we'll talk about them.", 'start': 1062.6, 'duration': 3.844}, {'end': 1068.405, 'text': 'All right.', 'start': 1067.885, 'duration': 0.52}, {'end': 1075.712, 'text': "so before we start talking about how we actually built that model and how it works, I guess I should convince you that I'm worth listening to.", 'start': 1068.405, 'duration': 7.307}], 'summary': 'Introduction to educational research tricks in course by dylan williams.', 'duration': 32.402, 'max_score': 1043.31, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/8SF_h3xF3cE/pics/8SF_h3xF3cE1043310.jpg'}], 'start': 597.628, 'title': 'Deep learning in art and language models', 'summary': "Highlights the fusion of deep learning with art, incorporating google's pathways language model to explain jokes, and emphasizes the importance of education in addressing ethical considerations in ai. additionally, it discusses an interactive classroom approach using colored cups for student feedback and an online version of the feedback system to facilitate effective learning.", 'chapters': [{'end': 748.652, 'start': 597.628, 'title': 'Deep learning in art and language models', 'summary': "Highlights the fusion of deep learning with art, including the use of google's pathways language model to explain jokes, as well as the ethical considerations in ai, emphasized by the importance of education in addressing these issues.", 'duration': 151.024, 'highlights': ['The fusion of deep learning and art is exemplified by fast.ai alumni with artistic backgrounds who have utilized deep learning to create innovative art, showcasing the potential of merging these fields.', "Google's pathways language model can generate answers and explanations for questions, including the ability to explain jokes, demonstrating the remarkable capabilities of language models in natural language processing.", 'The importance of ethical considerations in AI is emphasized, with a recommendation to explore the data ethics course at ethics.fast.ai, taught by Dr. Rachel Thomas, to delve deeper into these issues.', 'The significance of education in understanding and addressing ethical considerations in AI is highlighted, with a specific mention of Dylan Williams as a prominent figure in the field of education.', "The speaker, an AI researcher at the University of Queensland and fast.ai, also serves as a homeschooling primary school teacher, emphasizing the multifaceted nature of the individual's expertise."]}, {'end': 1081.816, 'start': 748.652, 'title': 'Interactive classroom approach', 'summary': 'Discusses an interactive classroom approach using colored cups for student feedback, an online version of the feedback system, and the unconventional teaching method to facilitate effective learning.', 'duration': 333.164, 'highlights': ["The teacher uses a colored cup system for student feedback, with green meaning 'fine,' yellow meaning 'not quite sure,' and red meaning 'no idea,' facilitating easy assessment of student understanding.", 'An online version of the colored cup system is utilized for remote learning, where students can click colored buttons to provide feedback, enabling the teacher to gauge student comprehension.', 'The teaching method focuses on starting with training a model rather than an in-depth review of linear algebra and calculus, aligning with research that emphasizes the effectiveness of learning with context in place.', 'The approach to learning deep learning is likened to the way people learn sports, gradually building skills and understanding, allowing students to become proficient at building and deploying models before delving into detailed technical aspects.', 'The course incorporates tricks from educational research literature to enhance the learning experience, aiming to make the learning process effective and natural for students.']}], 'duration': 484.188, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/8SF_h3xF3cE/pics/8SF_h3xF3cE597628.jpg', 'highlights': ["Google's pathways language model can explain jokes, showcasing remarkable language model capabilities.", 'The fusion of deep learning and art is exemplified by fast.ai alumni, demonstrating the potential of merging these fields.', 'The importance of education in addressing ethical considerations in AI is highlighted, emphasizing the multifaceted expertise of the speaker.', 'An online version of the colored cup system is utilized for remote learning, facilitating easy assessment of student understanding.', 'The teaching method focuses on starting with training a model, aligning with research emphasizing learning with context.', 'The course incorporates tricks from educational research literature to enhance the learning experience.']}, {'end': 1786.366, 'segs': [{'end': 1193.584, 'src': 'embed', 'start': 1123.285, 'weight': 1, 'content': [{'end': 1124.405, 'text': 'A lot of people like the book.', 'start': 1123.285, 'duration': 1.12}, {'end': 1127.868, 'text': 'Peter Norvig, director of research, loves the book.', 'start': 1125.966, 'duration': 1.902}, {'end': 1128.929, 'text': "In fact this one's here.", 'start': 1127.908, 'duration': 1.021}, {'end': 1132.751, 'text': 'One of the best sources for a program to become proficient in deep learning.', 'start': 1129.769, 'duration': 2.982}, {'end': 1136.554, 'text': 'Eric Topol loves the book.', 'start': 1135.533, 'duration': 1.021}, {'end': 1140.357, 'text': 'Hal Varian, emeritus professor at Berkeley, chief economist at Google, likes the book.', 'start': 1136.594, 'duration': 3.763}, {'end': 1145.521, 'text': 'Jerome Passenti, who is the head of AI at Facebook, likes the book.', 'start': 1141.458, 'duration': 4.063}, {'end': 1146.542, 'text': 'A lot of people like the book.', 'start': 1145.541, 'duration': 1.001}, {'end': 1151.927, 'text': "So hopefully you'll find that you like this material as well.", 'start': 1146.742, 'duration': 5.185}, {'end': 1157.929, 'text': "I've spent about 30 years of my life working in and around machine learning,", 'start': 1153.327, 'duration': 4.602}, {'end': 1167.893, 'text': 'including building a number of companies that relied on it and became the highest ranked competitor in the world on Kaggle in machine learning competitions.', 'start': 1157.929, 'duration': 9.964}, {'end': 1175.556, 'text': 'My company, Enlytic, which I founded, was the first company to specialize in deep learning for medicine,', 'start': 1169.574, 'duration': 5.982}, {'end': 1180.598, 'text': 'and MIT voted it one of the 50 smartest companies in 2016,, just above Facebook and SpaceX.', 'start': 1175.556, 'duration': 5.042}, {'end': 1193.584, 'text': "Um, I started Fast.ai with Rachel Thomas, and that was quite a few years ago now, but it's had a big impact on the world.", 'start': 1182.169, 'duration': 11.415}], 'summary': 'Many experts and industry leaders highly praise the book on deep learning, and the speaker has extensive experience in machine learning and deep learning, including building successful companies and being recognized as a top competitor in machine learning competitions.', 'duration': 70.299, 'max_score': 1123.285, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/8SF_h3xF3cE/pics/8SF_h3xF3cE1123285.jpg'}, {'end': 1254.095, 'src': 'embed', 'start': 1225.962, 'weight': 0, 'content': [{'end': 1232.467, 'text': 'NVIDIA started optimizing their stuff using our approaches, so it made quite a big difference there.', 'start': 1225.962, 'duration': 6.505}, {'end': 1238.749, 'text': "I'm the inventor of the ULM fit algorithm, which, according to the Transformers book,", 'start': 1234.468, 'duration': 4.281}, {'end': 1245.392, 'text': 'was one of the two key foundations behind the modern NLP revolution.', 'start': 1238.749, 'duration': 6.643}, {'end': 1254.095, 'text': 'This is the paper here and actually, you know, interesting point about that.', 'start': 1246.752, 'duration': 7.343}], 'summary': "Ulm fit algorithm revolutionized nlp, impacting nvidia's optimization.", 'duration': 28.133, 'max_score': 1225.962, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/8SF_h3xF3cE/pics/8SF_h3xF3cE1225962.jpg'}, {'end': 1563.458, 'src': 'embed', 'start': 1532.656, 'weight': 4, 'content': [{'end': 1538.439, 'text': 'And so remember, With a neural net, which is the basic function used in deep learning.', 'start': 1532.656, 'duration': 5.783}, {'end': 1542.401, 'text': "we don't have to hand code any of these or come up with any of these ideas.", 'start': 1538.439, 'duration': 3.962}, {'end': 1556.888, 'text': 'You just start with actually a random neural network and you feed it examples and you have it learn to recognize things and it turns out that these are the things that it creates for itself.', 'start': 1543.621, 'duration': 13.267}, {'end': 1563.458, 'text': 'so you can then combine these features.', 'start': 1558.556, 'duration': 4.902}], 'summary': 'Neural nets in deep learning learn to recognize features from examples.', 'duration': 30.802, 'max_score': 1532.656, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/8SF_h3xF3cE/pics/8SF_h3xF3cE1532656.jpg'}, {'end': 1655.575, 'src': 'embed', 'start': 1605.438, 'weight': 5, 'content': [{'end': 1612.904, 'text': "But this is the key difference of why we can now do things that previously we just didn't even conceive of as possible,", 'start': 1605.438, 'duration': 7.466}, {'end': 1617.227, 'text': "because now we don't have to hand code the features we look for.", 'start': 1612.904, 'duration': 4.323}, {'end': 1619.089, 'text': 'They can all be learned.', 'start': 1617.708, 'duration': 1.381}, {'end': 1632.078, 'text': "Now it's important to recognize we're going to be spending some time learning about building image-based algorithms,", 'start': 1621.631, 'duration': 10.447}, {'end': 1637.321, 'text': 'and image-based algorithms are not just for images, and in fact this is going to be a general theme.', 'start': 1632.078, 'duration': 5.243}, {'end': 1645.047, 'text': "We're going to show you some foundational techniques, but with creativity these foundational techniques can be used very widely.", 'start': 1637.481, 'duration': 7.566}, {'end': 1655.575, 'text': 'So for example, An image recognizer can also be used to classify sounds.', 'start': 1645.367, 'duration': 10.208}], 'summary': 'Advancements in machine learning enable diverse applications beyond images.', 'duration': 50.137, 'max_score': 1605.438, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/8SF_h3xF3cE/pics/8SF_h3xF3cE1605438.jpg'}, {'end': 1797.998, 'src': 'embed', 'start': 1759.62, 'weight': 6, 'content': [{'end': 1761.983, 'text': 'the myth that deep learning needs lots of data, I think,', 'start': 1759.62, 'duration': 2.363}, {'end': 1769.932, 'text': 'is mainly passed along by big companies that want to sell you computers to store lots of data and to process it.', 'start': 1761.983, 'duration': 7.949}, {'end': 1779.71, 'text': "We find that most real world projects don't need extraordinary amounts of data at, And, as you'll see,", 'start': 1770.793, 'duration': 8.917}, {'end': 1786.366, 'text': "there's actually a lot of fantastic places you can do state-of-the-art work for free nowadays, which is great news.", 'start': 1779.71, 'duration': 6.656}, {'end': 1797.998, 'text': "One of the key reasons for this is because of something called transfer learning, which we'll be learning about a lot during this course,", 'start': 1791.016, 'duration': 6.982}], 'summary': "Deep learning doesn't need lots of data; transfer learning enables state-of-the-art work for free.", 'duration': 38.378, 'max_score': 1759.62, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/8SF_h3xF3cE/pics/8SF_h3xF3cE1759620.jpg'}], 'start': 1083.237, 'title': 'Deep learning revolution', 'summary': "Discusses deep learning for coders, emphasizing the instructor's contributions to machine learning, including being the highest ranked kaggle competitor and founder of enlytic, and the evolution of image recognition, showcasing the shift to neural networks and their applications across domains with minimal resources.", 'chapters': [{'end': 1348.797, 'start': 1083.237, 'title': 'Deep learning for coders', 'summary': 'Outlines the background of the instructor, including his contributions to machine learning, such as being the highest ranked competitor in kaggle, the founder of enlytic, and the inventor of the ulm fit algorithm, and the significant impact of the fast.ai course in the industry and research.', 'duration': 265.56, 'highlights': ['Invented the ULM fit algorithm, a key foundation behind the modern NLP revolution, which has made a significant impact in the industry and research, including being used by Google and NVIDIA.', 'Founder of Enlytic, the first company specializing in deep learning for medicine, voted one of the 50 smartest companies by MIT in 2016, with global recognition for training big neural networks faster and cheaper than anyone in the world.', 'Highest ranked competitor in the world on Kaggle in machine learning competitions, with about 30 years of experience working in and around machine learning, and building companies relying on it.', "Co-author of the extremely popular book 'Deep Learning for Coders', heavily based on which the course is developed, with endorsements from prominent figures like Peter Norvig, Eric Topol, and Hal Varian."]}, {'end': 1786.366, 'start': 1348.877, 'title': 'Evolution of image recognition', 'summary': 'Explains the shift from classic machine learning approach to deep learning, highlighting the efficiency of neural networks in automatically learning features, with examples of its applications in various domains and the minimal resources required for real-world deep learning projects.', 'duration': 437.489, 'highlights': ["Neural networks automate feature extraction, eliminating the need for hand-coding, leading to the creation of complex feature detectors, as demonstrated by Zeiler and Fergus' study, revolutionizing the efficiency of image recognition models.", 'Deep learning techniques can be applied beyond images, such as classifying sounds, time series, and mouse movements, showcasing the versatility and creativity of foundational techniques in various domains.', 'Real-world deep learning projects often require minimal data and computational resources, debunking the myth that deep learning necessitates extensive data and expensive infrastructure, making state-of-the-art work accessible for free.', 'The shift from classic machine learning to deep learning has significantly reduced the complexity and time required for creating image recognition models, with neural networks automatically learning features, revolutionizing the efficiency of the process.']}], 'duration': 703.129, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/8SF_h3xF3cE/pics/8SF_h3xF3cE1083237.jpg', 'highlights': ['Invented ULM fit algorithm, key to NLP revolution, used by Google and NVIDIA', 'Founder of Enlytic, specializing in deep learning for medicine, recognized by MIT', 'Highest ranked Kaggle competitor in machine learning with 30 years of experience', "Co-author of 'Deep Learning for Coders', endorsed by Peter Norvig and others", 'Neural networks automate feature extraction, revolutionizing image recognition', 'Deep learning techniques applied beyond images, showcasing versatility', 'Real-world deep learning projects require minimal data and resources', 'Shift to deep learning reduced complexity and time for image recognition']}, {'end': 2158.337, 'segs': [{'end': 1871.273, 'src': 'embed', 'start': 1791.016, 'weight': 0, 'content': [{'end': 1797.998, 'text': "One of the key reasons for this is because of something called transfer learning, which we'll be learning about a lot during this course,", 'start': 1791.016, 'duration': 6.982}, {'end': 1801.399, 'text': "and it's something which very few people are aware of the power of.", 'start': 1797.998, 'duration': 3.401}, {'end': 1807.54, 'text': "In this course we'll be using PyTorch.", 'start': 1804.14, 'duration': 3.4}, {'end': 1817.043, 'text': 'For those of you who are not particularly close to the deep learning world, you might have heard of TensorFlow and not of PyTorch.', 'start': 1808.481, 'duration': 8.562}, {'end': 1830.755, 'text': 'You might be surprised to hear that TensorFlow has been dying in popularity in recent years and PyTorch is actually growing rapidly and in recent….', 'start': 1817.903, 'duration': 12.852}, {'end': 1839.06, 'text': 'research repositories amongst the top papers.', 'start': 1834.258, 'duration': 4.802}, {'end': 1843.662, 'text': 'TensorFlow is a tiny minority now compared to PyTorch.', 'start': 1840.6, 'duration': 3.062}, {'end': 1847.883, 'text': "This is also in great research that's come out from Ryan O'Connor.", 'start': 1844.682, 'duration': 3.201}, {'end': 1860.408, 'text': 'He also discovered that the majority of people that were doing TensorFlow in 2018, researchers, the majority have now shifted to PyTorch.', 'start': 1848.423, 'duration': 11.985}, {'end': 1868.051, 'text': "And I mention this because What people use in research is a very strong leading indicator of what's going to happen in industry,", 'start': 1860.828, 'duration': 7.223}, {'end': 1871.273, 'text': 'because this is where you know all the new algorithms are going to come out.', 'start': 1868.051, 'duration': 3.222}], 'summary': 'Pytorch is surpassing tensorflow in popularity, with majority of researchers shifting to pytorch.', 'duration': 80.257, 'max_score': 1791.016, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/8SF_h3xF3cE/pics/8SF_h3xF3cE1791016.jpg'}, {'end': 1999.82, 'src': 'embed', 'start': 1972.444, 'weight': 3, 'content': [{'end': 1978.306, 'text': "So when you use the code that we've provided for you, you know, you'll generally find you get better results.", 'start': 1972.444, 'duration': 5.862}, {'end': 1990.534, 'text': "So, fastai has been a really popular library and it's very widely used in industry, in academia and in teaching,", 'start': 1979.326, 'duration': 11.208}, {'end': 1999.82, 'text': "and as we go through this course we'll be seeing more and more pure PyTorch as we get deeper and deeper underneath to see exactly how things work.", 'start': 1990.534, 'duration': 9.286}], 'summary': 'Using provided code yields better results. fastai widely used in industry, academia, and teaching.', 'duration': 27.376, 'max_score': 1972.444, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/8SF_h3xF3cE/pics/8SF_h3xF3cE1972444.jpg'}, {'end': 2066.935, 'src': 'embed', 'start': 2039.547, 'weight': 4, 'content': [{'end': 2047.558, 'text': 'Jupyter notebook is the environment in which you will be doing most of your computing.', 'start': 2039.547, 'duration': 8.011}, {'end': 2062.393, 'text': "It's a web-based application which is extremely popular and widely used in industry and in academia and in teaching, and is a very, very,", 'start': 2048.98, 'duration': 13.413}, {'end': 2066.935, 'text': 'very powerful way to experiment and explore and to build.', 'start': 2062.393, 'duration': 4.542}], 'summary': 'Jupyter notebook is a widely used web-based application for computing and experimentation.', 'duration': 27.388, 'max_score': 2039.547, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/8SF_h3xF3cE/pics/8SF_h3xF3cE2039547.jpg'}, {'end': 2135.656, 'src': 'embed', 'start': 2105.377, 'weight': 5, 'content': [{'end': 2109.203, 'text': "So Kaggle doesn't just have competitions, but it also has a cloud notebook server.", 'start': 2105.377, 'duration': 3.826}, {'end': 2115.071, 'text': "And I've got quite a few examples there.", 'start': 2110.164, 'duration': 4.907}, {'end': 2127.651, 'text': 'So let me give you a quick example of how we use Jupyter notebooks to build stuff, to experiment, to explore.', 'start': 2119.045, 'duration': 8.606}, {'end': 2135.656, 'text': "So on Kaggle, if you start with somebody else's notebook, so why don't you start with this one, Jupyter notebook 101.", 'start': 2128.731, 'duration': 6.925}], 'summary': 'Kaggle offers a cloud notebook server, with diverse examples for jupyter notebooks.', 'duration': 30.279, 'max_score': 2105.377, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/8SF_h3xF3cE/pics/8SF_h3xF3cE2105377.jpg'}], 'start': 1791.016, 'title': "Pytorch's dominance", 'summary': "Discusses pytorch's increasing popularity over tensorflow, with research repositories showing a significant shift towards pytorch, and the advantages of using fastai library built on top of pytorch and jupyter notebooks in computing, particularly on cloud servers like kaggle.", 'chapters': [{'end': 1889.422, 'start': 1791.016, 'title': "Pytorch's growing popularity", 'summary': 'Discusses the increasing popularity of pytorch over tensorflow in recent years, with research repositories showing a significant shift towards pytorch and a notable increase in industry adoption.', 'duration': 98.406, 'highlights': ['PyTorch is rapidly growing in popularity, with research repositories showing a majority shift from TensorFlow, indicating a strong leading indicator of industry adoption.', 'The majority of researchers who were using TensorFlow in 2018 have now shifted to PyTorch, signifying a significant shift in industry preference.', 'Transfer learning and the technical fundamentals of PyTorch are highlighted as key reasons for its increasing adoption over TensorFlow.']}, {'end': 2158.337, 'start': 1890.686, 'title': 'Fastai and jupyter notebooks', 'summary': 'Discusses the advantages of using the fastai library built on top of pytorch, highlighting the reduction in code and better results, and the significance of using jupyter notebooks in computing, particularly on cloud servers like kaggle, showcasing its popularity and powerful capabilities.', 'duration': 267.651, 'highlights': ['The fastai library provides a significant reduction in code compared to plain PyTorch, resulting in better results and following best practices.', 'Jupyter notebooks are widely used in industry, academia, and teaching, offering a powerful environment for experimenting and building.', "Kaggle provides a cloud notebook server and allows users to copy and edit others' notebooks, encouraging upvoting to support creators and sharing."]}], 'duration': 367.321, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/8SF_h3xF3cE/pics/8SF_h3xF3cE1791016.jpg', 'highlights': ['PyTorch is rapidly growing in popularity, with research repositories showing a majority shift from TensorFlow, indicating a strong leading indicator of industry adoption.', 'The majority of researchers who were using TensorFlow in 2018 have now shifted to PyTorch, signifying a significant shift in industry preference.', 'Transfer learning and the technical fundamentals of PyTorch are highlighted as key reasons for its increasing adoption over TensorFlow.', 'The fastai library provides a significant reduction in code compared to plain PyTorch, resulting in better results and following best practices.', 'Jupyter notebooks are widely used in industry, academia, and teaching, offering a powerful environment for experimenting and building.', "Kaggle provides a cloud notebook server and allows users to copy and edit others' notebooks, encouraging upvoting to support creators and sharing."]}, {'end': 2427.918, 'segs': [{'end': 2221.213, 'src': 'embed', 'start': 2159.217, 'weight': 0, 'content': [{'end': 2164.481, 'text': 'And to use it we can type in any arbitrary expression in Python and click run.', 'start': 2159.217, 'duration': 5.264}, {'end': 2168.522, 'text': 'And the very first time we do that, it says session is starting.', 'start': 2166.095, 'duration': 2.427}, {'end': 2172.432, 'text': "It's basically launching a virtual computer for us to run our code.", 'start': 2168.562, 'duration': 3.87}, {'end': 2173.455, 'text': 'This is all free.', 'start': 2172.913, 'duration': 0.542}, {'end': 2179.304, 'text': "In a sense it's like the world's most powerful calculator.", 'start': 2175.942, 'duration': 3.362}, {'end': 2187.789, 'text': "It's a calculator where you have all of the capabilities of the world's, I think, most popular programming language.", 'start': 2180.345, 'duration': 7.444}, {'end': 2192.291, 'text': 'certainly it and JavaScript would be the top two directly at your disposal.', 'start': 2187.789, 'duration': 4.502}, {'end': 2197.074, 'text': 'So Python does know how to do one plus one and so you can see here it spits out the answer.', 'start': 2192.671, 'duration': 4.403}, {'end': 2204.798, 'text': 'I hate clicking, I always use keyboard shortcuts, so instead of clicking this little arrow you just press shift enter to do the same thing.', 'start': 2198.134, 'duration': 6.664}, {'end': 2211.507, 'text': "as you can see there's not just calculations here, there's also pros.", 'start': 2206.864, 'duration': 4.643}, {'end': 2221.213, 'text': "And so Jupyter notebooks are great for explaining to you the version of yourself in six months time what on earth you're doing, or to your co-workers,", 'start': 2212.327, 'duration': 8.886}], 'summary': 'Jupyter notebooks provide a free, powerful platform for running python code and explaining complex concepts, with capabilities of the most popular programming languages directly accessible.', 'duration': 61.996, 'max_score': 2159.217, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/8SF_h3xF3cE/pics/8SF_h3xF3cE2159217.jpg'}, {'end': 2392.195, 'src': 'heatmap', 'start': 2273.363, 'weight': 5, 'content': [{'end': 2279.567, 'text': "That's not Python, that's a bash shell command, okay, so that's what the exclamation mark means.", 'start': 2273.363, 'duration': 6.204}, {'end': 2285.753, 'text': 'As you can see, you can put images into notebooks,', 'start': 2282.891, 'duration': 2.862}, {'end': 2291.535, 'text': 'and so the image I popped in here was the one showing that Jupyter won the 2017 software system award,', 'start': 2285.753, 'duration': 5.782}, {'end': 2295.837, 'text': 'which is pretty much the biggest award there is for this kind of software.', 'start': 2291.535, 'duration': 4.302}, {'end': 2300.699, 'text': "Okay so that's the basic idea of how we use notebooks.", 'start': 2296.737, 'duration': 3.962}, {'end': 2310.323, 'text': "So let's have a look at how we do our bird or not bird model.", 'start': 2306.061, 'duration': 4.262}, {'end': 2325.879, 'text': "One thing I always like to do when I'm using something like Colab or Kaggle cloud platforms that I'm not controlling is make sure that I'm using the most recent version of any software.", 'start': 2316.291, 'duration': 9.588}, {'end': 2333.726, 'text': 'So my first cell here is exclamation mark pip install minus u, that means upgrade, q for quiet, fastai.', 'start': 2326.56, 'duration': 7.166}, {'end': 2337.529, 'text': 'so that makes sure that we have the latest version of fastai.', 'start': 2334.526, 'duration': 3.003}, {'end': 2339.61, 'text': 'and if you always have that at the start of your notebooks,', 'start': 2337.529, 'duration': 2.081}, {'end': 2343.774, 'text': "you're never going to have those awkward foreign threads where you say why isn't this working?", 'start': 2339.61, 'duration': 4.164}, {'end': 2349.839, 'text': "and somebody says to you oh, you're using an old version of some software.", 'start': 2343.774, 'duration': 6.065}, {'end': 2361.989, 'text': "so you'll see here, this notebook is the exact thing that i was showing you at the start of this lesson.", 'start': 2349.839, 'duration': 12.15}, {'end': 2373.78, 'text': "so If you haven't done much Python, you might be surprised about how little code there is here.", 'start': 2361.989, 'duration': 11.791}, {'end': 2380.726, 'text': 'And so Python is a concise but not too concise language.', 'start': 2374.641, 'duration': 6.085}, {'end': 2384.949, 'text': "You'll see that there's less boilerplate than some other languages you might be familiar with.", 'start': 2381.546, 'duration': 3.403}, {'end': 2389.493, 'text': "And I'm also taking advantage of a lot of libraries.", 'start': 2386.751, 'duration': 2.742}, {'end': 2392.195, 'text': 'So fastai provides a lot of convenient things for you.', 'start': 2389.553, 'duration': 2.642}], 'summary': 'Jupyter won the 2017 software system award, demonstrating its significance in the field of software. upgrading to the latest version of fastai is recommended to avoid compatibility issues.', 'duration': 60.363, 'max_score': 2273.363, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/8SF_h3xF3cE/pics/8SF_h3xF3cE2273363.jpg'}, {'end': 2427.918, 'src': 'embed', 'start': 2393.616, 'weight': 6, 'content': [{'end': 2410.03, 'text': 'um so i forgot to import um so To use a external library we use import, to import a symbol from a library.', 'start': 2393.616, 'duration': 16.414}, {'end': 2413.492, 'text': 'Fast.ai has a lot of libraries we provide.', 'start': 2411.311, 'duration': 2.181}, {'end': 2420.535, 'text': 'They generally start with fast something, so for example, to make it easy to download a URL, fast download has download URL.', 'start': 2413.732, 'duration': 6.803}, {'end': 2427.918, 'text': 'To make it easy to create a thumbnail, we have image to thumb, and so forth.', 'start': 2422.016, 'duration': 5.902}], 'summary': 'Fast.ai offers a variety of libraries, such as fast download for url and image to thumb for thumbnail creation.', 'duration': 34.302, 'max_score': 2393.616, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/8SF_h3xF3cE/pics/8SF_h3xF3cE2393616.jpg'}], 'start': 2159.217, 'title': 'Python as a powerful calculator and jupyter notebooks for efficient coding', 'summary': "Explains how python serves as a powerful calculator and introduces the efficient use of jupyter notebooks, covering keyboard shortcuts, code and markdown cells, bash shell commands, image insertion, and ensuring software's latest version, with a focus on fastai library and its features.", 'chapters': [{'end': 2197.074, 'start': 2159.217, 'title': 'Python as a powerful calculator', 'summary': "Explains how python can be used as a powerful calculator, providing access to the capabilities of the world's most popular programming languages, and launching a virtual computer for running code.", 'duration': 37.857, 'highlights': ["Python provides access to the capabilities of the world's most popular programming languages, making it a powerful calculator.", 'Launching a virtual computer for running code is a free service provided by Python.', "Python can perform basic arithmetic operations like 'one plus one' and display the result."]}, {'end': 2427.918, 'start': 2198.134, 'title': 'Jupyter notebooks for fast and efficient coding', 'summary': 'Introduces the efficient use of jupyter notebooks, including keyboard shortcuts, creating cells for code and markdown, using bash shell commands, inserting images, and ensuring the latest version of software is installed, with a focus on fastai library and its convenient features.', 'duration': 229.784, 'highlights': ['Jupyter notebooks are great for explaining to your future self, co-workers, and the open source community, as well as for blogging purposes.', 'Using keyboard shortcuts like shift enter can streamline tasks and avoid excessive clicking.', "Ensuring the latest version of software is installed using commands like 'pip install -u' can prevent issues related to outdated software versions.", 'Fastai library offers convenient features like easy URL download and image thumbnail creation.', 'Jupyter won the 2017 software system award, highlighting its significance in the software community.']}], 'duration': 268.701, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/8SF_h3xF3cE/pics/8SF_h3xF3cE2159217.jpg', 'highlights': ["Python provides access to the capabilities of the world's most popular programming languages, making it a powerful calculator.", 'Launching a virtual computer for running code is a free service provided by Python.', "Python can perform basic arithmetic operations like 'one plus one' and display the result.", 'Jupyter notebooks are great for explaining to your future self, co-workers, and the open source community, as well as for blogging purposes.', 'Using keyboard shortcuts like shift enter can streamline tasks and avoid excessive clicking.', "Ensuring the latest version of software is installed using commands like 'pip install -u' can prevent issues related to outdated software versions.", 'Fastai library offers convenient features like easy URL download and image thumbnail creation.', 'Jupyter won the 2017 software system award, highlighting its significance in the software community.']}, {'end': 2958.527, 'segs': [{'end': 2496.673, 'src': 'embed', 'start': 2466.025, 'weight': 1, 'content': [{'end': 2467.287, 'text': 'And it does that in parallel.', 'start': 2466.025, 'duration': 1.262}, {'end': 2470.711, 'text': 'So it does that, you know, surprisingly quickly.', 'start': 2467.927, 'duration': 2.784}, {'end': 2475.319, 'text': "One other fastai thing I'm using here is resize images.", 'start': 2472.977, 'duration': 2.342}, {'end': 2483.304, 'text': "You generally will find that for computer vision algorithms you don't need particularly big images.", 'start': 2477.64, 'duration': 5.664}, {'end': 2489.848, 'text': "so I'm resizing these to a maximum sight length of 400, because it's actually much faster.", 'start': 2483.304, 'duration': 6.544}, {'end': 2496.673, 'text': 'Because GPUs are so quick, for big images most of the time can be taken up just opening it.', 'start': 2490.549, 'duration': 6.124}], 'summary': 'Fastai quickly resizes images to max length of 400, saving time for computer vision algorithms.', 'duration': 30.648, 'max_score': 2466.025, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/8SF_h3xF3cE/pics/8SF_h3xF3cE2466025.jpg'}, {'end': 2538.563, 'src': 'embed', 'start': 2511.403, 'weight': 0, 'content': [{'end': 2519.59, 'text': "So the data block is the key thing that you're going to want to get familiar with as deep learning practitioners at the start of your journey,", 'start': 2511.403, 'duration': 8.187}, {'end': 2527.595, 'text': "because the main thing you're going to be trying to figure out is how do I get this data into my model?", 'start': 2519.59, 'duration': 8.005}, {'end': 2529.657, 'text': 'Now that might surprise you.', 'start': 2528.496, 'duration': 1.161}, {'end': 2538.563, 'text': 'You might be thinking we should be spending all of our time talking about neural network architectures and matrix multiplication and gradients and stuff like that.', 'start': 2529.797, 'duration': 8.766}], 'summary': 'Understanding the data block is crucial for deep learning practitioners to efficiently input data into models at the start of their journey.', 'duration': 27.16, 'max_score': 2511.403, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/8SF_h3xF3cE/pics/8SF_h3xF3cE2511403.jpg'}, {'end': 2572.769, 'src': 'embed', 'start': 2548.129, 'weight': 4, 'content': [{'end': 2560.379, 'text': "the deep learningist community has found a reasonably small number of types of model that work for nearly all the main applications you'll need,", 'start': 2548.129, 'duration': 12.25}, {'end': 2565.903, 'text': 'and fastai will create the right type of model for you the vast majority of the time.', 'start': 2560.379, 'duration': 5.524}, {'end': 2572.769, 'text': 'So all of that stuff about tweaking neural network architectures and stuff.', 'start': 2567.164, 'duration': 5.605}], 'summary': 'Deep learningist community has found a small number of models that work for most applications; fastai creates the right model most of the time.', 'duration': 24.64, 'max_score': 2548.129, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/8SF_h3xF3cE/pics/8SF_h3xF3cE2548129.jpg'}, {'end': 2839.914, 'src': 'heatmap', 'start': 2668.631, 'weight': 3, 'content': [{'end': 2678.075, 'text': 'So the first thing that we tell fastai is, what kind of input do we have? And so there are lots of blocks in fast.ai for different kinds of inputs.', 'start': 2668.631, 'duration': 9.444}, {'end': 2679.436, 'text': 'So we said, oh, the input is an image.', 'start': 2678.095, 'duration': 1.341}, {'end': 2684.358, 'text': "What kind of output is there? What kind of label? The output's a category.", 'start': 2681.057, 'duration': 3.301}, {'end': 2687.68, 'text': "So that means it's one of a number of possibilities.", 'start': 2684.578, 'duration': 3.102}, {'end': 2693.583, 'text': "So that's enough for fast.ai to know what kind of model to build for you.", 'start': 2689.821, 'duration': 3.762}, {'end': 2701.424, 'text': 'So what are the items in this model? What am I actually going to be looking at to train from? this is a function.', 'start': 2695.764, 'duration': 5.66}, {'end': 2706.489, 'text': 'In fact you might have noticed, if you were looking carefully, that we use this function here.', 'start': 2702.265, 'duration': 4.224}, {'end': 2715.138, 'text': "It's a function which returns a list of all of the image files in a path, based on extension.", 'start': 2708.291, 'duration': 6.847}, {'end': 2721.424, 'text': "So, every time it's going to try and find out what things to train from, it's going to use that function, which, in this case,", 'start': 2716.639, 'duration': 4.785}, {'end': 2722.445, 'text': 'will get a list of image files.', 'start': 2721.424, 'duration': 1.021}, {'end': 2732.008, 'text': "Now, something we'll talk about shortly is that it's critical that you put aside some data for testing the accuracy of your model,", 'start': 2724.505, 'duration': 7.503}, {'end': 2733.809, 'text': "and that's called a validation set.", 'start': 2732.008, 'duration': 1.801}, {'end': 2738.971, 'text': "It's so critical that fastai won't let you train a model without one.", 'start': 2735.41, 'duration': 3.561}, {'end': 2750.156, 'text': 'So you actually have to tell it how to create a validation set, how to set aside some data and in this case we say randomly set aside 20% of the data.', 'start': 2740.812, 'duration': 9.344}, {'end': 2761.633, 'text': 'Okay, next question, then you have to tell fastai, is how do we know the correct label of a photo?', 'start': 2753.45, 'duration': 8.183}, {'end': 2764.274, 'text': "How do we know if it's a bird photo or a forest photo?", 'start': 2761.793, 'duration': 2.481}, {'end': 2777.403, 'text': 'And this is another function, and this function simply returns the parent folder of a path, and so in this case,', 'start': 2765.755, 'duration': 11.648}, {'end': 2782.989, 'text': 'we saved our images into either forest or bird.', 'start': 2777.403, 'duration': 5.586}, {'end': 2784.411, 'text': "so that's where the labels are going to come from.", 'start': 2782.989, 'duration': 1.422}, {'end': 2796.006, 'text': 'And then finally, Most computer vision architectures need all of your inputs as you train to be the same size.', 'start': 2786.914, 'duration': 9.092}, {'end': 2803.97, 'text': 'So item transforms are all of the bits of code that are going to run on every item.', 'start': 2797.406, 'duration': 6.564}, {'end': 2813.558, 'text': "on every image in this case, and we're saying okay we want you to resize each of them to being 192 by 192 pixels.", 'start': 2804.87, 'duration': 8.688}, {'end': 2823.407, 'text': "There's two ways you can resize, you can either crop out a piece in the middle or you can squish it, and so we're saying squish it.", 'start': 2814.599, 'duration': 8.808}, {'end': 2832.509, 'text': "So that's the data block, that's all that you need, and from there we create an important class called data loaders.", 'start': 2825.785, 'duration': 6.724}, {'end': 2839.914, 'text': 'Data loaders are the things that actually PyTorch iterates through to grab a bunch of your data at a time.', 'start': 2833.05, 'duration': 6.864}], 'summary': 'Fast.ai automates model building by specifying input, output, validation, label extraction, and item transformation, creating data loaders.', 'duration': 24.952, 'max_score': 2668.631, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/8SF_h3xF3cE/pics/8SF_h3xF3cE2668631.jpg'}, {'end': 2750.156, 'src': 'embed', 'start': 2716.639, 'weight': 2, 'content': [{'end': 2721.424, 'text': "So, every time it's going to try and find out what things to train from, it's going to use that function, which, in this case,", 'start': 2716.639, 'duration': 4.785}, {'end': 2722.445, 'text': 'will get a list of image files.', 'start': 2721.424, 'duration': 1.021}, {'end': 2732.008, 'text': "Now, something we'll talk about shortly is that it's critical that you put aside some data for testing the accuracy of your model,", 'start': 2724.505, 'duration': 7.503}, {'end': 2733.809, 'text': "and that's called a validation set.", 'start': 2732.008, 'duration': 1.801}, {'end': 2738.971, 'text': "It's so critical that fastai won't let you train a model without one.", 'start': 2735.41, 'duration': 3.561}, {'end': 2750.156, 'text': 'So you actually have to tell it how to create a validation set, how to set aside some data and in this case we say randomly set aside 20% of the data.', 'start': 2740.812, 'duration': 9.344}], 'summary': 'Fastai requires a validation set, randomly setting aside 20% of the data for testing the accuracy of the model.', 'duration': 33.517, 'max_score': 2716.639, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/8SF_h3xF3cE/pics/8SF_h3xF3cE2716639.jpg'}, {'end': 2848.059, 'src': 'embed', 'start': 2797.406, 'weight': 6, 'content': [{'end': 2803.97, 'text': 'So item transforms are all of the bits of code that are going to run on every item.', 'start': 2797.406, 'duration': 6.564}, {'end': 2813.558, 'text': "on every image in this case, and we're saying okay we want you to resize each of them to being 192 by 192 pixels.", 'start': 2804.87, 'duration': 8.688}, {'end': 2823.407, 'text': "There's two ways you can resize, you can either crop out a piece in the middle or you can squish it, and so we're saying squish it.", 'start': 2814.599, 'duration': 8.808}, {'end': 2832.509, 'text': "So that's the data block, that's all that you need, and from there we create an important class called data loaders.", 'start': 2825.785, 'duration': 6.724}, {'end': 2839.914, 'text': 'Data loaders are the things that actually PyTorch iterates through to grab a bunch of your data at a time.', 'start': 2833.05, 'duration': 6.864}, {'end': 2848.059, 'text': 'The way it can do it so fast is by using a GPU, which is something that can do thousands of things at the same time,', 'start': 2840.994, 'duration': 7.065}], 'summary': 'Using item transforms to resize images to 192x192 pixels and create efficient data loaders for pytorch by utilizing gpu for fast processing.', 'duration': 50.653, 'max_score': 2797.406, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/8SF_h3xF3cE/pics/8SF_h3xF3cE2797406.jpg'}], 'start': 2430.347, 'title': 'Data preparation and data block in fast.ai', 'summary': 'Discusses the significance of visualizing and preprocessing data for deep learning models, highlighting the usage of the data block command and the fastai library. it also emphasizes the key components of the data block in fast.ai and the practical importance of functional programming style.', 'chapters': [{'end': 2593.619, 'start': 2430.347, 'title': 'Data preparation for deep learning', 'summary': 'Discusses the importance of visualizing and preprocessing data for deep learning models, emphasizing the usage of the data block command and the fastai library for efficient image handling and model creation.', 'duration': 163.272, 'highlights': ['The data block command is emphasized as a crucial tool for deep learning practitioners to efficiently integrate data into models, reducing the need for extensive focus on neural network architectures and matrix operations.', 'The fastai library enables efficient image handling by providing features such as parallel downloading of images and resizing to a maximum length of 400, significantly enhancing processing speed.', 'Visual inspection of data at every step is highlighted as a critical practice, ensuring the reasonableness of acquired bird and forest photos before downloading, contributing to data quality and model performance.']}, {'end': 2958.527, 'start': 2594.42, 'title': 'Data block in fast.ai', 'summary': 'Discusses the key components of the data block in fast.ai, including the input and output types, item transforms, validation set creation, and the creation of data loaders, emphasizing the practical importance of functional programming style.', 'duration': 364.107, 'highlights': ['The data block in Fast.ai is designed to handle different types of datasets by focusing on key components such as input type, output type, item transforms, validation set creation, and data loaders.', 'Functional style programming is highlighted as a practical approach for the work in Fast.ai, with the usage of functional-style methods like map for efficient implementation.', 'The creation of a validation set is emphasized as critical, with Fast.ai requiring a validation set to train a model, exemplified by setting aside 20% of the data for validation.', 'The importance of data loaders in Fast.ai, which iterates through and feeds a batch of images to the training algorithm using a GPU for faster processing, is discussed.', 'The significance of item transforms is explained, with a specific example of resizing images to 192x192 pixels using the squishing method, and its role in preparing inputs for computer vision architectures.']}], 'duration': 528.18, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/8SF_h3xF3cE/pics/8SF_h3xF3cE2430347.jpg', 'highlights': ['The data block command is crucial for integrating data into models, reducing focus on neural network architectures.', 'Fastai library enables efficient image handling with parallel downloading and resizing to a maximum length of 400.', 'Visual inspection of data at every step is critical for ensuring data quality and model performance.', 'The data block in Fast.ai handles different datasets by focusing on key components like input type and output type.', 'Functional style programming is practical for work in Fast.ai, using methods like map for efficient implementation.', 'Creation of a validation set is critical, with Fast.ai requiring a validation set to train a model.', 'Data loaders in Fast.ai iterate through and feed a batch of images to the training algorithm using a GPU for faster processing.', 'Item transforms play a significant role in preparing inputs for computer vision architectures.']}, {'end': 3858.741, 'segs': [{'end': 2990.858, 'src': 'embed', 'start': 2958.527, 'weight': 3, 'content': [{'end': 2975.391, 'text': "we've got an object called dls that stand for data loaders and that contains iterators that PyTorch can run through to grab batches of randomly split out training images to train the model with,", 'start': 2958.527, 'duration': 16.864}, {'end': 2977.331, 'text': 'and validation images to test the model with.', 'start': 2975.391, 'duration': 1.94}, {'end': 2980.432, 'text': 'So now we need a model.', 'start': 2979.692, 'duration': 0.74}, {'end': 2985.673, 'text': 'The critical concept here in fast.ai is called a learner.', 'start': 2982.75, 'duration': 2.923}, {'end': 2990.858, 'text': 'A learner is something which combines a model, which is, that is,', 'start': 2986.413, 'duration': 4.445}], 'summary': "Using dls, pytorch grabs batches of training and validation images to train and test the model. fast.ai's critical concept is a learner.", 'duration': 32.331, 'max_score': 2958.527, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/8SF_h3xF3cE/pics/8SF_h3xF3cE2958527.jpg'}, {'end': 3039.433, 'src': 'embed', 'start': 3013.253, 'weight': 0, 'content': [{'end': 3019.139, 'text': "there's a relatively small number that basically work for the vast majority of things you do.", 'start': 3013.253, 'duration': 5.886}, {'end': 3027.247, 'text': "If you pass in just a bear symbol like this, it's going to be one of fastai's built-in models.", 'start': 3021.661, 'duration': 5.586}, {'end': 3036.731, 'text': "but what's particularly interesting is that we integrate a wonderful library by Ross Whiteman called TIM, the PyTorch Image Models,", 'start': 3027.247, 'duration': 9.484}, {'end': 3039.433, 'text': 'which is the largest collection of computer vision models in the world.', 'start': 3036.731, 'duration': 2.702}], 'summary': "Fastai's built-in models and integration with tim, the largest collection of computer vision models in the world, offer a wide range of options for various tasks.", 'duration': 26.18, 'max_score': 3013.253, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/8SF_h3xF3cE/pics/8SF_h3xF3cE3013253.jpg'}, {'end': 3183.734, 'src': 'embed', 'start': 3158.251, 'weight': 1, 'content': [{'end': 3163.393, 'text': "And so then, something that fastai has that's unique is this fine-tune method,", 'start': 3158.251, 'duration': 5.142}, {'end': 3180.241, 'text': 'which what it does is it takes those pre-trained weights we downloaded for you and it adjusts them in a really carefully controlled way to just teach the model the differences between your data set and what it was originally trained for.', 'start': 3163.393, 'duration': 16.848}, {'end': 3181.982, 'text': "That's called fine-tuning.", 'start': 3181.001, 'duration': 0.981}, {'end': 3183.734, 'text': 'hence the name.', 'start': 3183.294, 'duration': 0.44}], 'summary': "Fastai's unique fine-tune method adjusts pre-trained weights for better model performance.", 'duration': 25.483, 'max_score': 3158.251, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/8SF_h3xF3cE/pics/8SF_h3xF3cE3158251.jpg'}, {'end': 3524.162, 'src': 'embed', 'start': 3486.443, 'weight': 7, 'content': [{'end': 3489.05, 'text': "We create something called a U-Net learner, which we'll learn about later.", 'start': 3486.443, 'duration': 2.607}, {'end': 3491.436, 'text': 'And then again we call fine-tune.', 'start': 3489.691, 'duration': 1.745}, {'end': 3496.873, 'text': "So that is it and that's how we create a segmentation model.", 'start': 3492.87, 'duration': 4.003}, {'end': 3507.722, 'text': 'What about stepping away from computer vision? So perhaps the most widely used kind of model used in industry is tabular analysis.', 'start': 3499.315, 'duration': 8.407}, {'end': 3511.645, 'text': 'So taking things like spreadsheets and database tables and trying to predict columns of those.', 'start': 3507.762, 'duration': 3.883}, {'end': 3524.162, 'text': "So in tabular analysis it really looks very similar to what we've seen already.", 'start': 3517.719, 'duration': 6.443}], 'summary': 'Introducing u-net learner for segmentation model and discussing tabular analysis for predicting columns.', 'duration': 37.719, 'max_score': 3486.443, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/8SF_h3xF3cE/pics/8SF_h3xF3cE3486443.jpg'}, {'end': 3587.214, 'src': 'heatmap', 'start': 3529.065, 'weight': 0.71, 'content': [{'end': 3536.089, 'text': "this is the thing in fastai that downloads some data and decompresses it for you, and there's a whole lot of URLs provided by fastai for.", 'start': 3529.065, 'duration': 7.024}, {'end': 3540.313, 'text': 'So all the kind of common data sets that you might want to use,', 'start': 3536.949, 'duration': 3.364}, {'end': 3545.439, 'text': 'you know all the ones that are in the book or lots of data sets that are kind of widely used in learning and research.', 'start': 3540.313, 'duration': 5.126}, {'end': 3547.862, 'text': 'So that makes life nice and easy for you.', 'start': 3545.459, 'duration': 2.403}, {'end': 3552.047, 'text': "So again we're going to create data loaders, but this time it's tabular data loaders.", 'start': 3548.863, 'duration': 3.184}, {'end': 3555.731, 'text': 'But we provide pretty similar kind of information to what we had before.', 'start': 3552.468, 'duration': 3.263}, {'end': 3557.835, 'text': 'A couple of new things.', 'start': 3556.593, 'duration': 1.242}, {'end': 3565.749, 'text': 'we have to tell it which of the columns are categorical, so they can only take one of a few values, and which ones are continuous,', 'start': 3557.835, 'duration': 7.914}, {'end': 3568.253, 'text': 'so they can take basically any real number.', 'start': 3565.749, 'duration': 2.504}, {'end': 3576.092, 'text': "And then again we can use the exact same show batch that we've seen before to see the data.", 'start': 3569.97, 'duration': 6.122}, {'end': 3587.214, 'text': "And so fastai uses a lot of something called type dispatch, which is a system that's particularly popular in, in a language called Julia,", 'start': 3576.892, 'duration': 10.322}], 'summary': 'Fastai provides easy access to common data sets, creating tabular data loaders for categorical and continuous columns with type dispatch system.', 'duration': 58.149, 'max_score': 3529.065, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/8SF_h3xF3cE/pics/8SF_h3xF3cE3529065.jpg'}, {'end': 3665.022, 'src': 'embed', 'start': 3632.151, 'weight': 4, 'content': [{'end': 3633.353, 'text': "In this case it's a tabular learner.", 'start': 3632.151, 'duration': 1.202}, {'end': 3639.749, 'text': "Now this time we don't say fine-tune, we say fit, specifically fit one cycle.", 'start': 3635.287, 'duration': 4.462}, {'end': 3647.133, 'text': "That's because for tabular models there's not generally going to be a pre-trained model that already does something like what you want,", 'start': 3640.81, 'duration': 6.323}, {'end': 3650.114, 'text': 'because every table of data is very different.', 'start': 3647.133, 'duration': 2.981}, {'end': 3657.818, 'text': "Whereas pictures often have a similar theme, you know, they're all pictures, they all have the same kind of general idea of what pictures are.", 'start': 3650.955, 'duration': 6.863}, {'end': 3665.022, 'text': "So that's why it generally doesn't make too much sense to fine-tune a tabular model, so instead you just fit.", 'start': 3658.879, 'duration': 6.143}], 'summary': "For tabular models, it's more effective to fit one cycle instead of fine-tuning due to the diverse nature of table data.", 'duration': 32.871, 'max_score': 3632.151, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/8SF_h3xF3cE/pics/8SF_h3xF3cE3632151.jpg'}, {'end': 3712.656, 'src': 'embed', 'start': 3680.25, 'weight': 5, 'content': [{'end': 3684.133, 'text': 'Collaborative filtering is the basis of most recommendation systems today.', 'start': 3680.25, 'duration': 3.883}, {'end': 3695.761, 'text': "It's a system where we basically take data that says which users liked which products or which users used which products,", 'start': 3684.613, 'duration': 11.148}, {'end': 3703.906, 'text': 'and then we use that to guess what other products those users might like, based on finding similar users and what those similar users liked.', 'start': 3695.761, 'duration': 8.145}, {'end': 3712.656, 'text': "The interesting thing about collaborative filtering is that when we say similar users, we're not referring to similar demographically,", 'start': 3704.967, 'duration': 7.689}], 'summary': 'Collaborative filtering uses user data to recommend similar products.', 'duration': 32.406, 'max_score': 3680.25, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/8SF_h3xF3cE/pics/8SF_h3xF3cE3680250.jpg'}, {'end': 3810.44, 'src': 'embed', 'start': 3755.567, 'weight': 2, 'content': [{'end': 3761.491, 'text': 'in this case we can just say from csv and pass in a csv and this is what collaborative filtering data looks like.', 'start': 3755.567, 'duration': 5.924}, {'end': 3769.316, 'text': "it's going to have, generally speaking, a user id, some kind of product id, in this case a movie and a rating.", 'start': 3761.491, 'duration': 7.825}, {'end': 3778.682, 'text': 'So in this case this user gave this movie a rating of 3.5 out of 5.', 'start': 3771.637, 'duration': 7.045}, {'end': 3780.383, 'text': 'And so again, you can see show batch right.', 'start': 3778.682, 'duration': 1.701}, {'end': 3781.284, 'text': 'so use show batch.', 'start': 3780.383, 'duration': 0.901}, {'end': 3786.687, 'text': 'you should get back some useful visualization of your data, regardless of what kind of data it is.', 'start': 3781.284, 'duration': 5.403}, {'end': 3798.616, 'text': "And so again we create a learner, this time it's the collaborative filtering learner and you pass in your data.", 'start': 3790.59, 'duration': 8.026}, {'end': 3808.88, 'text': "In this case, we give it one extra piece of information, which is because this is not predicting a category, but it's predicting a real number.", 'start': 3800.817, 'duration': 8.063}, {'end': 3810.44, 'text': "we tell it what's the possible range.", 'start': 3808.88, 'duration': 1.56}], 'summary': 'Using collaborative filtering data with user id, product id, and ratings, a learner is created to predict real numbers within a specified range.', 'duration': 54.873, 'max_score': 3755.567, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/8SF_h3xF3cE/pics/8SF_h3xF3cE3755567.jpg'}], 'start': 2958.527, 'title': 'Pytorch data loaders, fast.ai learner, image models, and model creation', 'summary': 'Introduces pytorch data loaders and fast.ai learner, explores pytorch image models integration in fast.ai, and covers tabular and collaborative filtering models with unique fine-tuning methods, providing comprehensive insights into computer vision and recommendation systems.', 'chapters': [{'end': 3013.253, 'start': 2958.527, 'title': 'Pytorch data loaders and fast.ai learner', 'summary': 'Introduces the concept of data loaders in pytorch, which contain iterators for grabbing batches of training and validation images, and emphasizes the critical concept of a learner in fast.ai, which combines a model and data for training.', 'duration': 54.726, 'highlights': ['The critical concept here in fast.ai is called a learner, which combines a model and data for training.', 'An object called dls contains iterators for grabbing batches of training and validation images in PyTorch.']}, {'end': 3599.197, 'start': 3013.253, 'title': 'Fastai: pytorch image models and model creation', 'summary': 'Explores the integration of tim, the pytorch image models, by fast.ai, which includes the largest collection of computer vision models in the world, and the unique fine-tune method for pre-trained models, demonstrating the creation of computer vision, segmentation, and tabular analysis models using fast.ai.', 'duration': 585.944, 'highlights': ['fastai integrates TIM, the PyTorch Image Models, the largest collection of computer vision models in the world.', 'Fast.ai is the first and only framework to integrate TIM, allowing the use of any PyTorch image model.', 'The unique fine-tune method adjusts pre-trained weights for the purpose of recognizing specific data sets, resulting in 100% accuracy after a few seconds.', 'Creation of computer vision, segmentation, and tabular analysis models using fast.ai, demonstrating the versatility and ease of model creation.']}, {'end': 3858.741, 'start': 3599.797, 'title': 'Tabular and collaborative filtering models', 'summary': 'Covers building a tabular model for salary data and creating a collaborative filtering model for recommendation systems. it emphasizes the difference in fine-tuning approaches between tabular and collaborative filtering models and highlights the concept and application of collaborative filtering.', 'duration': 258.944, 'highlights': ['Creating a tabular learner model for predicting salary based on demographic data, with no need for fine-tuning due to the unique nature of tabular data.', 'Explaining collaborative filtering as the basis of most recommendation systems, using data on user-product interactions to predict user preferences, and highlighting the distinction of similarity based on product preferences rather than demographics.', 'Demonstrating the creation of collaborative filtering data loaders from a csv file, showcasing the structure of the data including user id, product id, and rating, and highlighting the use of show batch for data visualization.', 'Training a collaborative filtering learner model to predict real number ratings within a specified range, and showcasing the mean squared error as a measure of model performance during training.']}], 'duration': 900.214, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/8SF_h3xF3cE/pics/8SF_h3xF3cE2958527.jpg', 'highlights': ['Fast.ai integrates TIM, the PyTorch Image Models, the largest collection of computer vision models in the world.', 'The unique fine-tune method adjusts pre-trained weights for the purpose of recognizing specific data sets, resulting in 100% accuracy after a few seconds.', 'Demonstrating the creation of collaborative filtering data loaders from a csv file, showcasing the structure of the data including user id, product id, and rating, and highlighting the use of show batch for data visualization.', 'An object called dls contains iterators for grabbing batches of training and validation images in PyTorch.', 'Creating a tabular learner model for predicting salary based on demographic data, with no need for fine-tuning due to the unique nature of tabular data.', 'Explaining collaborative filtering as the basis of most recommendation systems, using data on user-product interactions to predict user preferences, and highlighting the distinction of similarity based on product preferences rather than demographics.', 'The critical concept here in fast.ai is called a learner, which combines a model and data for training.', 'Creation of computer vision, segmentation, and tabular analysis models using fast.ai, demonstrating the versatility and ease of model creation.', 'Training a collaborative filtering learner model to predict real number ratings within a specified range, and showcasing the mean squared error as a measure of model performance during training.']}, {'end': 4971.984, 'segs': [{'end': 4042.106, 'src': 'embed', 'start': 3979.26, 'weight': 1, 'content': [{'end': 3992.747, 'text': 'So you might be surprised to discover that if you go to fastai slash fastai that the source code for the entire library is notebooks.', 'start': 3979.26, 'duration': 13.487}, {'end': 4006.818, 'text': "And so the nice thing about this is that you know the source code for the Fast.io library has actual pictures of the actual things that we're building,", 'start': 3998.792, 'duration': 8.026}, {'end': 4007.358, 'text': 'for example.', 'start': 4006.818, 'duration': 0.54}, {'end': 4015.904, 'text': 'What else have we done with notebooks? Oh, blogging.', 'start': 4011.341, 'duration': 4.563}, {'end': 4028.817, 'text': 'I love blogging with notebooks because when I want to explain something I just write the code and you can just see the outputs and it all just works.', 'start': 4018.45, 'duration': 10.367}, {'end': 4039.564, 'text': 'Another thing you might be surprised by is all of our tests and continuous integration are also all in notebooks.', 'start': 4033.02, 'duration': 6.544}, {'end': 4042.106, 'text': 'So, every time we change one of our notebooks….', 'start': 4040.204, 'duration': 1.902}], 'summary': "Fast.ai's entire library source code is in notebooks, including tests and continuous integration.", 'duration': 62.846, 'max_score': 3979.26, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/8SF_h3xF3cE/pics/8SF_h3xF3cE3979260.jpg'}, {'end': 4126.109, 'src': 'embed', 'start': 4077.788, 'weight': 6, 'content': [{'end': 4087.816, 'text': "at present? We're still scratching the tip of the iceberg, even though it's a pretty well-hyped, you know, heavily marketed technology at this point.", 'start': 4077.788, 'duration': 10.028}, {'end': 4094.702, 'text': 'You know, when we started in 2014, or so you know,', 'start': 4087.896, 'duration': 6.806}, {'end': 4101.147, 'text': 'not many people were talking about deep learning and really there was no accessible way to get started with it.', 'start': 4094.702, 'duration': 6.445}, {'end': 4103.368, 'text': 'There were no pre-trained models you could download.', 'start': 4101.167, 'duration': 2.201}, {'end': 4113.823, 'text': 'you know, there was just starting to appear some of the first open source software that would run on GPUs.', 'start': 4104.899, 'duration': 8.924}, {'end': 4114.584, 'text': 'but yeah, I mean.', 'start': 4113.823, 'duration': 0.761}, {'end': 4118.526, 'text': "but despite the fact that today there's a lot of people talking about deep learning, we're just scratching the surface.", 'start': 4114.584, 'duration': 3.942}, {'end': 4126.109, 'text': 'Every time pretty much somebody says to me I work in domain x and I thought I might try deep learning out to see if it can help.', 'start': 4119.627, 'duration': 6.482}], 'summary': "Deep learning was not accessible in 2014, now heavily marketed, but we're still just scratching the surface.", 'duration': 48.321, 'max_score': 4077.788, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/8SF_h3xF3cE/pics/8SF_h3xF3cE4077788.jpg'}, {'end': 4205.067, 'src': 'embed', 'start': 4142.607, 'weight': 0, 'content': [{'end': 4157.091, 'text': 'So in NLP, deep learning is the state-of-the-art method in all these kinds of things and a lot more computer vision, medicine, biology,', 'start': 4142.607, 'duration': 14.484}, {'end': 4164.258, 'text': 'recommendation systems, playing games, robotics.', 'start': 4161.215, 'duration': 3.043}, {'end': 4171.406, 'text': "I mean it's just… I've tried elsewhere to make bigger lists and I just end up with pages and pages and pages.", 'start': 4165.279, 'duration': 6.127}, {'end': 4176.792, 'text': "So you know it's….", 'start': 4171.886, 'duration': 4.906}, {'end': 4177.813, 'text': 'generally speaking, you know.', 'start': 4176.792, 'duration': 1.021}, {'end': 4189.399, 'text': "if it's something that a human can do reasonably quickly, like look at a go board and decide if it looks like a good go board or not,", 'start': 4177.813, 'duration': 11.586}, {'end': 4197.022, 'text': "even if it needs to be an expert human, then that's probably something that deep learning will be pretty good at.", 'start': 4189.399, 'duration': 7.623}, {'end': 4205.067, 'text': "if it's something that takes a lot of logical thought processes over an extended period of time, you know,", 'start': 4198.482, 'duration': 6.585}], 'summary': 'Deep learning is state-of-the-art in nlp, computer vision, medicine, recommendation systems, games, and robotics.', 'duration': 62.46, 'max_score': 4142.607, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/8SF_h3xF3cE/pics/8SF_h3xF3cE4142607.jpg'}, {'end': 4266.648, 'src': 'embed', 'start': 4231.979, 'weight': 9, 'content': [{'end': 4233.421, 'text': 'This was the first neural network.', 'start': 4231.979, 'duration': 1.442}, {'end': 4235.402, 'text': 'Remember, neural networks are the basis of deep learning.', 'start': 4233.461, 'duration': 1.941}, {'end': 4241.367, 'text': 'So this was back in 1957.', 'start': 4236.483, 'duration': 4.884}, {'end': 4255.078, 'text': 'The basic ideas have not changed much at all, but you know we do have things like GPUs now and solid-state drives and stuff like that,', 'start': 4241.367, 'duration': 13.711}, {'end': 4258.741, 'text': 'and of course much more data just is available now.', 'start': 4255.078, 'duration': 3.663}, {'end': 4266.648, 'text': 'but this has been decades of really hard work by a lot of people to get to this point.', 'start': 4258.741, 'duration': 7.907}], 'summary': 'The first neural network, basis of deep learning, dates back to 1957, with advancements in technology and availability of more data.', 'duration': 34.669, 'max_score': 4231.979, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/8SF_h3xF3cE/pics/8SF_h3xF3cE4231979.jpg'}, {'end': 4700.576, 'src': 'embed', 'start': 4667.334, 'weight': 5, 'content': [{'end': 4676.918, 'text': 'Deploying machine learning models in practice can come with a lot of know, little tricky details,', 'start': 4667.334, 'duration': 9.584}, {'end': 4682.683, 'text': "but the basic idea in your code is that you're just going to have a line of code that says learn.predict,", 'start': 4676.918, 'duration': 5.765}, {'end': 4685.125, 'text': 'and then you just fit it in with all the rest of your code in the usual way.', 'start': 4682.683, 'duration': 2.442}, {'end': 4691.051, 'text': 'And this is why, because a trained model is just another thing that maps inputs to results.', 'start': 4685.586, 'duration': 5.465}, {'end': 4700.576, 'text': 'Okay, All right.', 'start': 4691.071, 'duration': 9.505}], 'summary': 'Deploying machine learning models involves fitting a line of code (learn.predict) with the rest of the code, as a trained model maps inputs to results.', 'duration': 33.242, 'max_score': 4667.334, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/8SF_h3xF3cE/pics/8SF_h3xF3cE4667334.jpg'}, {'end': 4760.026, 'src': 'embed', 'start': 4726.438, 'weight': 7, 'content': [{'end': 4728.38, 'text': "There's obviously a lot you're going to have to learn.", 'start': 4726.438, 'duration': 1.942}, {'end': 4739.412, 'text': "And to be clear, I'm not going to be teaching Python in this course, but we do have links to great Python resources in the forum,", 'start': 4729.161, 'duration': 10.251}, {'end': 4741.334, 'text': 'so check out that thread.', 'start': 4739.412, 'duration': 1.922}, {'end': 4748.419, 'text': "regardless of where you're at, the most important thing is to experiment.", 'start': 4743.636, 'duration': 4.783}, {'end': 4760.026, 'text': "And so experimenting could be as simple as just running those Kaggle notebooks that I've shown you, just to see them run.", 'start': 4749.66, 'duration': 10.366}], 'summary': 'Course emphasizes learning through experimentation with python resources available in the forum.', 'duration': 33.588, 'max_score': 4726.438, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/8SF_h3xF3cE/pics/8SF_h3xF3cE4726438.jpg'}], 'start': 3858.741, 'title': 'Deep learning applications and machine learning basics', 'summary': 'Discusses the benefits of using jupyter notebooks, the broad applicability of deep learning in nlp, computer vision, medicine, biology, and recommendation systems, emphasizing tasks suitable for deep learning. it also explains the basics of machine learning, including loss calculation, weight updates, model flexibility, and practical use of trained models, highlighting the importance of experimentation and sharing findings in the forum.', 'chapters': [{'end': 4141.826, 'start': 3858.741, 'title': 'Deep learning progress and notebook benefits', 'summary': 'Discusses the benefits of using jupyter notebooks, the rise of deep learning, and the fact that despite the widespread interest in deep learning, there is still untapped potential, with many fields yet to benefit from its capabilities.', 'duration': 283.085, 'highlights': ['Jupyter notebooks are extensively used for various tasks including writing books, creating plots, training models, blogging, and running tests and continuous integration.', 'Deep learning is still in its early stages despite being heavily marketed, with many fields yet to benefit from its capabilities.', 'Many people have achieved state-of-the-art results in their respective fields by applying deep learning, indicating untapped potential in various domains.', 'The source code for the entire fastai library is written in notebooks, making it accessible and providing actual pictures of the things being built.', 'The rise of deep learning has seen a surge in interest, but there is still a vast amount of untapped potential in various domains.']}, {'end': 4454.946, 'start': 4142.607, 'title': 'Deep learning in nlp', 'summary': 'Discusses the broad applicability of deep learning in nlp, computer vision, medicine, biology, recommendation systems, playing games, and robotics, emphasizing that tasks requiring quick human-like decisions are suitable for deep learning, and highlights the historical development of neural networks and the basic idea of machine learning.', 'duration': 312.339, 'highlights': ["Deep learning's broad applicability in NLP, computer vision, medicine, biology, recommendation systems, playing games, and robotics.", 'Tasks suitable for deep learning are those requiring quick human-like decisions and logical thought processes over an extended period of time.', 'Historical development of neural networks and the basic idea of machine learning.']}, {'end': 4971.984, 'start': 4455.166, 'title': 'Understanding machine learning basics', 'summary': 'Explains the basics of machine learning, including concepts such as loss calculation, weight updates, model flexibility, and the practical use of trained models, emphasizing the importance of experimentation and sharing findings in the forum.', 'duration': 516.818, 'highlights': ['The chapter explains the basics of machine learning, including concepts such as loss calculation, weight updates, model flexibility, and the practical use of trained models.', 'The chapter emphasizes the importance of experimentation and sharing findings in the forum, encouraging learners to try different exercises and push themselves, regardless of their familiarity with Python.']}], 'duration': 1113.243, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/8SF_h3xF3cE/pics/8SF_h3xF3cE3858741.jpg', 'highlights': ["Deep learning's broad applicability in NLP, computer vision, medicine, biology, recommendation systems, playing games, and robotics.", 'Jupyter notebooks are extensively used for various tasks including writing books, creating plots, training models, blogging, and running tests and continuous integration.', 'Many people have achieved state-of-the-art results in their respective fields by applying deep learning, indicating untapped potential in various domains.', 'The source code for the entire fastai library is written in notebooks, making it accessible and providing actual pictures of the things being built.', 'Tasks suitable for deep learning are those requiring quick human-like decisions and logical thought processes over an extended period of time.', 'The chapter explains the basics of machine learning, including concepts such as loss calculation, weight updates, model flexibility, and the practical use of trained models.', 'The rise of deep learning has seen a surge in interest, but there is still a vast amount of untapped potential in various domains.', 'The chapter emphasizes the importance of experimentation and sharing findings in the forum, encouraging learners to try different exercises and push themselves, regardless of their familiarity with Python.', 'Deep learning is still in its early stages despite being heavily marketed, with many fields yet to benefit from its capabilities.', 'Historical development of neural networks and the basic idea of machine learning.']}], 'highlights': ['Creating a computer vision model to classify 200 pictures of birds and 200 pictures of forests, running on a laptop in under 30 seconds', 'Invented ULM fit algorithm, key to NLP revolution, used by Google and NVIDIA', 'PyTorch is rapidly growing in popularity, with research repositories showing a majority shift from TensorFlow, indicating a strong leading indicator of industry adoption', "Python provides access to the capabilities of the world's most popular programming languages, making it a powerful calculator", 'The data block command is crucial for integrating data into models, reducing focus on neural network architectures', "Deep learning's broad applicability in NLP, computer vision, medicine, biology, recommendation systems, playing games, and robotics", 'Jupyter notebooks are extensively used for various tasks including writing books, creating plots, training models, blogging, and running tests and continuous integration', 'Fast.ai integrates TIM, the PyTorch Image Models, the largest collection of computer vision models in the world']}