title
Lesson 1 - Deep Learning for Coders (2020)

description
Welcome to Deep Learning for Coders! Be sure to watch these videos through https://course.fast.ai to get access to the searchable transcript, interactive notebooks, setup guides, questionnaires, and so forth. We don't recommend watching the videos directly on YouTube. In this first lesson, we learn about what deep learning is, and how it's connected to machine learning, and regular computer programming. We get our GPU-powered deep learning server set up, and use it to train models across vision, NLP, tabular data, and collaborative filtering. We do this all in Jupyter Notebooks, using transfer learning from pretrained models for the vision and NLP training. We discuss the important topics of test and validation sets, and how to create and use them to avoid over-fitting. We learn about some key jargon used in deep learning. We also discuss how AI projects can fail, and techniques for avoiding failure. 00:00 - Introduction 06:44 - What you don’t need to do deep learning 08:38 - What is the point of learning deep learning 09:52 - Neural Nets: a brief history 16:00 - Top to bottom learning approach 23:06 - The software stack 39:06 - Git Repositories 42:20 - First practical exercise in Jupyter Notebook 48:00 - Interpretation and explanation of the exercise 55:35 - Stochastic Gradient Descent (SGD) 1:01:30 - Consider how a model interacts with its environment 1:07:42 - "doc" function and fastai framework documentation 1:16:20 - Image Segmentation 1:17:34 - Classifying a review's sentiment based on IMDB text reviews 1:18:30 - Predicting salary based on tabular data from CSV 1:20:15 - Lesson Summary

detail
{'title': 'Lesson 1 - Deep Learning for Coders (2020)', 'heatmap': [{'end': 1881.538, 'start': 1783.995, 'weight': 0.978}], 'summary': '《lesson 1 - deep learning for coders (2020)》 covers the live launch of fast ai library version 2 during a pandemic, outlines the fast book syllabus, introduces fast.ai, data ethics, deep learning fundamentals, jupyter notebook features, machine learning, neural network architecture basics, impact of biased data on predictive policing, and python programming for data science, providing valuable insights and practical guidance.', 'chapters': [{'end': 259.832, 'segs': [{'end': 105.346, 'src': 'embed', 'start': 0.652, 'weight': 0, 'content': [{'end': 6.516, 'text': 'So hello everybody and welcome to Deep Learning for Coders Lesson 1.', 'start': 0.652, 'duration': 5.864}, {'end': 17.744, 'text': "This is the fourth year that we've done this, but it's a very different and very special version for a number of reasons.", 'start': 6.516, 'duration': 11.228}, {'end': 26.41, 'text': "The first reason it's different is because we are bringing it to you live from day number one of a complete shutdown or not complete shutdown,", 'start': 17.884, 'duration': 8.526}, {'end': 28.311, 'text': 'but nearly complete shutdown of San Francisco.', 'start': 26.41, 'duration': 1.901}, {'end': 34.012, 'text': "We're going to be recording it over the next two months in the midst of this global pandemic.", 'start': 29.411, 'duration': 4.601}, {'end': 41.615, 'text': "So if things seem a little crazy sometimes in this course, I apologize, but that's why this is happening.", 'start': 34.333, 'duration': 7.282}, {'end': 53.447, 'text': "The other reason it's special is because We're trying to make this our kind of definitive version, right?", 'start': 43.355, 'duration': 10.092}, {'end': 60.509, 'text': "Since we've been doing this for a while now, we've actually finally gotten to the point where we almost feel like we know what we're talking about.", 'start': 54.267, 'duration': 6.242}, {'end': 72.092, 'text': "To the point that Sylvain and I have actually written a book and we've actually written a piece of software from scratch called the Fast AI Library Version 2..", 'start': 61.349, 'duration': 10.743}, {'end': 77.234, 'text': "We've written a peer-reviewed paper about this library.", 'start': 72.092, 'duration': 5.142}, {'end': 87.674, 'text': 'So this is kind of designed to be like the version of the course that is Hopefully going to last a while.', 'start': 78.125, 'duration': 9.549}, {'end': 91.738, 'text': 'The syllabus is based very closely on this book, right?', 'start': 87.674, 'duration': 4.064}, {'end': 98.745, 'text': 'So if you want to read along properly as you go, Please buy it.', 'start': 92.239, 'duration': 6.506}, {'end': 105.346, 'text': 'and I say please buy it because actually the whole thing is also available for free in the form of Jupyter notebooks.', 'start': 98.745, 'duration': 6.601}], 'summary': 'Deep learning for coders lesson 1 being live from san francisco amid pandemic, with a new book and software.', 'duration': 104.694, 'max_score': 0.652, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/_QUEXsHfsA0/pics/_QUEXsHfsA0652.jpg'}, {'end': 169.796, 'src': 'embed', 'start': 138.466, 'weight': 6, 'content': [{'end': 148.95, 'text': 'So we have a big request here, which is The deal is this You can read this thing for free as Jupyter notebooks,', 'start': 138.466, 'duration': 10.484}, {'end': 156.353, 'text': 'but That is not as convenient as reading it on a Kindle or, you know, in a paper book or whatever.', 'start': 148.95, 'duration': 7.403}, {'end': 158.674, 'text': "So please don't turn this into a PDF.", 'start': 156.353, 'duration': 2.321}, {'end': 169.796, 'text': "Please don't turn it into a form designed more for reading, because the whole point is that we hope that you'll buy it.", 'start': 159.566, 'duration': 10.23}], 'summary': 'Request to not convert free content into pdf to encourage purchase.', 'duration': 31.33, 'max_score': 138.466, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/_QUEXsHfsA0/pics/_QUEXsHfsA0138466.jpg'}, {'end': 236.505, 'src': 'embed', 'start': 197.049, 'weight': 4, 'content': [{'end': 201.632, 'text': "it's not nice and don't be that person.", 'start': 197.049, 'duration': 4.583}, {'end': 206.996, 'text': 'So either way, you can read along with the syllabus in the book.', 'start': 201.772, 'duration': 5.224}, {'end': 213.981, 'text': "There's a couple of different versions of of these notebooks, right?", 'start': 208.858, 'duration': 5.123}, {'end': 218.421, 'text': "There's the.", 'start': 214.581, 'duration': 3.84}, {'end': 226.343, 'text': "there's the full notebook that has the entire prose pictures everything.", 'start': 218.421, 'duration': 7.922}, {'end': 236.505, 'text': 'Now we actually wrote a system to turn notebooks into a printed book and sometimes that looks kind of weird.', 'start': 227.463, 'duration': 9.042}], 'summary': 'Discussion on various versions of notebooks and system for turning them into printed books.', 'duration': 39.456, 'max_score': 197.049, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/_QUEXsHfsA0/pics/_QUEXsHfsA0197049.jpg'}], 'start': 0.652, 'title': 'Learning and syllabus', 'summary': 'Discusses the live launch of deep learning for coders lesson 1 during a pandemic, development of fast ai library version 2, and a peer-reviewed paper. it also outlines the fast book syllabus, available as jupyter notebooks for free, with a request not to convert it into a pdf, and details the different versions of the notebooks and their peculiarities.', 'chapters': [{'end': 77.234, 'start': 0.652, 'title': 'Deep learning for coders lesson 1', 'summary': 'Discusses the special version of deep learning for coders lesson 1 being brought live from a complete shutdown of san francisco due to a global pandemic, along with the development of fast ai library version 2 and a peer-reviewed paper.', 'duration': 76.582, 'highlights': ['The chapter discusses the special version of Deep Learning for Coders Lesson 1 being brought live from a complete shutdown of San Francisco due to a global pandemic, along with the development of Fast AI Library Version 2 and a peer-reviewed paper.', 'The course is being recorded over the next two months in the midst of the global pandemic.', 'Sylvain and the speaker have written a book and developed Fast AI Library Version 2, along with a peer-reviewed paper about this library.', 'The course is in its fourth year, signifying its long-standing nature.']}, {'end': 259.832, 'start': 78.125, 'title': 'Fast book syllabus and jupyter notebooks', 'summary': "Discusses the fast book syllabus closely based on a book, also available for free as jupyter notebooks thanks to o'reilly media's generosity, with a request not to convert it into a pdf, and outlines the different versions of the notebooks and their peculiarities.", 'duration': 181.707, 'highlights': ["The syllabus is closely based on a book and is also available for free as Jupyter notebooks, thanks to O'Reilly media's generosity.", "There's a request not to convert the Jupyter notebooks into a PDF to encourage people to buy the book.", 'Different versions of the notebooks are available, including a full notebook with prose, pictures, and everything, and a version designed to turn into a printed book with added information.', 'The chapter emphasizes the need for readers to buy the book rather than turning the free Jupyter notebooks into a PDF.']}], 'duration': 259.18, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/_QUEXsHfsA0/pics/_QUEXsHfsA0652.jpg', 'highlights': ['The live launch of Deep Learning for Coders Lesson 1 during a pandemic', 'Development of Fast AI Library Version 2 and a peer-reviewed paper', 'The course being recorded over the next two months in the midst of the global pandemic', 'The syllabus is closely based on a book and is available for free as Jupyter notebooks', 'Different versions of the notebooks are available, including a full notebook with prose, pictures, and everything', 'The course is in its fourth year, signifying its long-standing nature', 'A request not to convert the Jupyter notebooks into a PDF to encourage people to buy the book']}, {'end': 895.1, 'segs': [{'end': 295.113, 'src': 'embed', 'start': 261.493, 'weight': 0, 'content': [{'end': 270.199, 'text': 'Now, when I say we, who is we? Well, I mentioned, one important part of the we is Sylvain.', 'start': 261.493, 'duration': 8.706}, {'end': 275.882, 'text': 'Sylvain is my co-author of the book and the Fast.ai version 2 library.', 'start': 270.819, 'duration': 5.063}, {'end': 278.463, 'text': 'So he is my partner in crime here.', 'start': 276.002, 'duration': 2.461}, {'end': 284.727, 'text': 'The other key we here is Rachel Thomas.', 'start': 279.344, 'duration': 5.383}, {'end': 288.509, 'text': 'And so maybe Rachel you can come and say hello.', 'start': 285.547, 'duration': 2.962}, {'end': 289.97, 'text': "She's the co-founder of Fast.ai.", 'start': 288.569, 'duration': 1.401}, {'end': 295.113, 'text': "Hello Yes, I'm the co-founder of FastAI.", 'start': 292.452, 'duration': 2.661}], 'summary': 'Sylvain is a co-author of the book and fast.ai version 2 library, rachel thomas is the co-founder of fast.ai.', 'duration': 33.62, 'max_score': 261.493, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/_QUEXsHfsA0/pics/_QUEXsHfsA0261493.jpg'}, {'end': 384.489, 'src': 'embed', 'start': 353.615, 'weight': 2, 'content': [{'end': 356.376, 'text': 'She is the founding director of the Center for Applied Data Ethics.', 'start': 353.615, 'duration': 2.761}, {'end': 360.377, 'text': 'At the University of San Francisco.', 'start': 357.016, 'duration': 3.361}, {'end': 361.818, 'text': 'At the University of San Francisco.', 'start': 360.517, 'duration': 1.301}, {'end': 362.118, 'text': 'Thank you.', 'start': 361.838, 'duration': 0.28}, {'end': 369.362, 'text': "We're going to be talking about data ethics throughout the course because, well, we happen to think it's very important.", 'start': 363.098, 'duration': 6.264}, {'end': 378.306, 'text': "So for those parts, although I'll generally be presenting them, they will be on the whole, based on Rachel's work,", 'start': 370.222, 'duration': 8.084}, {'end': 380.247, 'text': "because she actually knows what she's talking about.", 'start': 378.306, 'duration': 1.941}, {'end': 384.489, 'text': "Although thanks to her I kind of know a bit about what I'm talking about too.", 'start': 381.768, 'duration': 2.721}], 'summary': 'Rachel is the founding director of the center for applied data ethics at the university of san francisco, emphasizing the importance of data ethics.', 'duration': 30.874, 'max_score': 353.615, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/_QUEXsHfsA0/pics/_QUEXsHfsA0353615.jpg'}, {'end': 433.037, 'src': 'embed', 'start': 409.927, 'weight': 4, 'content': [{'end': 419.091, 'text': "you attempting to learn, uh, deep learning, or are you, um, too stupid, or you don't have, uh, enough vast, uh, resources or whatever?", 'start': 409.927, 'duration': 9.164}, {'end': 421.012, 'text': "Because that's what a lot of people are telling us.", 'start': 419.211, 'duration': 1.801}, {'end': 426.014, 'text': "They're saying you need teams of PhDs and massive data centers full of GPUs.", 'start': 421.072, 'duration': 4.942}, {'end': 428.235, 'text': "Otherwise, um, it's, it's pointless.", 'start': 426.094, 'duration': 2.141}, {'end': 429.955, 'text': "Um, don't worry.", 'start': 428.955, 'duration': 1}, {'end': 431.696, 'text': 'That is not at all true.', 'start': 430.115, 'duration': 1.581}, {'end': 433.037, 'text': "It couldn't be further from the truth.", 'start': 431.796, 'duration': 1.241}], 'summary': "Learning deep learning doesn't require teams of phds or massive data centers.", 'duration': 23.11, 'max_score': 409.927, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/_QUEXsHfsA0/pics/_QUEXsHfsA0409927.jpg'}, {'end': 578.634, 'src': 'embed', 'start': 552.688, 'weight': 1, 'content': [{'end': 557.972, 'text': 'So it is not speculative at this point whether this is a useful tool.', 'start': 552.688, 'duration': 5.284}, {'end': 561.295, 'text': "It's a useful tool in lots and lots and lots of places.", 'start': 558.012, 'duration': 3.283}, {'end': 562.456, 'text': 'Extremely useful tool.', 'start': 561.495, 'duration': 0.961}, {'end': 569.112, 'text': 'And in many of these cases it is equivalent to, or better than, human performance,', 'start': 563.056, 'duration': 6.056}, {'end': 575.193, 'text': 'at least according to some particular kind of narrow definition of things that humans do in these kinds of areas.', 'start': 569.112, 'duration': 6.081}, {'end': 578.634, 'text': 'So deep learning is pretty amazing.', 'start': 576.173, 'duration': 2.461}], 'summary': 'Deep learning is an extremely useful tool, often equivalent to or better than human performance in many cases.', 'duration': 25.946, 'max_score': 552.688, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/_QUEXsHfsA0/pics/_QUEXsHfsA0552688.jpg'}, {'end': 848.779, 'src': 'embed', 'start': 820.037, 'weight': 3, 'content': [{'end': 830.747, 'text': "So in the 80s, during and after this was released, people started building in this second layer of neurons, avoiding Minsky's problem.", 'start': 820.037, 'duration': 10.71}, {'end': 840.597, 'text': 'And in fact, it was shown that it was mathematically provable that by adding that one extra layer of neurons,', 'start': 831.568, 'duration': 9.029}, {'end': 848.779, 'text': 'it was enough to allow any mathematical model to be approximated to any level of accuracy with these neural networks.', 'start': 840.597, 'duration': 8.182}], 'summary': "In the 80s, adding a second layer of neurons solved minsky's problem, allowing any mathematical model to be approximated with neural networks.", 'duration': 28.742, 'max_score': 820.037, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/_QUEXsHfsA0/pics/_QUEXsHfsA0820037.jpg'}, {'end': 906.589, 'src': 'embed', 'start': 870.827, 'weight': 5, 'content': [{'end': 874.888, 'text': 'I was using them for the very boring things like targeted marketing for retail banks.', 'start': 870.827, 'duration': 4.061}, {'end': 879.069, 'text': 'They tended to be big companies with lots of money that were using them.', 'start': 875.488, 'duration': 3.581}, {'end': 886.454, 'text': 'And it certainly though was true that often the networks were too big or slow to be useful.', 'start': 880.23, 'duration': 6.224}, {'end': 895.1, 'text': 'They were certainly useful for some things, but they, you know, they never felt to me like they were living up to the promise for some reason.', 'start': 887.015, 'duration': 8.085}, {'end': 906.589, 'text': "Now, what I didn't know and nobody I personally met knew was that actually there were researchers that had showed 30 years ago that to get practical good performance,", 'start': 895.761, 'duration': 10.828}], 'summary': 'Ai not living up to promise for targeted marketing, big companies using it, but often too big or slow to be useful.', 'duration': 35.762, 'max_score': 870.827, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/_QUEXsHfsA0/pics/_QUEXsHfsA0870827.jpg'}], 'start': 261.493, 'title': 'Fast.ai, data ethics, and power of deep learning', 'summary': 'Introduces fast.ai and data ethics, challenging myths about deep learning. it also explores the history and wide applicability of deep learning, showcasing its superiority in various industries and tasks.', 'chapters': [{'end': 517.383, 'start': 261.493, 'title': 'Introduction to fast.ai and data ethics', 'summary': 'Introduces the co-authors of the fast.ai library, emphasizing the importance of data ethics and debunking the myth that deep learning requires vast resources and technical expertise.', 'duration': 255.89, 'highlights': ['Sylvain and Rachel are co-authors of the Fast.ai library and are experts in math and data ethics respectively, providing valuable insights throughout the course.', 'Rachel Thomas, the co-founder of Fast.ai, is also the founding director of the Center for Applied Data Ethics at the University of San Francisco, bringing real-world expertise in data ethics.', 'The course dispels the myth that deep learning requires extensive resources and technical expertise, highlighting that world-class research and industry projects have been achieved using minimal resources and expertise.']}, {'end': 895.1, 'start': 520.958, 'title': 'The power of deep learning', 'summary': 'Explores the wide applicability and effectiveness of deep learning, tracing its origins from neural networks in the 1940s to the breakthroughs in the 1980s, and its current impact across various industries, showcasing its superiority in many tasks compared to human performance.', 'duration': 374.142, 'highlights': ['Deep learning is a useful tool in many areas and is equivalent to, or better than, human performance in certain tasks, showcasing its wide applicability and effectiveness. Demonstrated superiority over human performance in various tasks.', 'The breakthrough in the 1980s showed that adding a second layer of neurons made it mathematically provable that any mathematical model could be approximated to any level of accuracy with neural networks, signifying a significant advancement in the capabilities of neural networks. Mathematically proven capability to approximate any mathematical model to any level of accuracy with neural networks.', 'Neural networks have been widely used in industry since the early 90s, particularly in big companies for tasks such as targeted marketing for retail banks, demonstrating their early adoption and impact in various industries. Widespread adoption in industry, particularly in big companies for various applications.']}], 'duration': 633.607, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/_QUEXsHfsA0/pics/_QUEXsHfsA0261493.jpg', 'highlights': ['Sylvain and Rachel are co-authors of the Fast.ai library and are experts in math and data ethics respectively, providing valuable insights throughout the course.', 'Deep learning is a useful tool in many areas and is equivalent to, or better than, human performance in certain tasks, showcasing its wide applicability and effectiveness. Demonstrated superiority over human performance in various tasks.', 'Rachel Thomas, the co-founder of Fast.ai, is also the founding director of the Center for Applied Data Ethics at the University of San Francisco, bringing real-world expertise in data ethics.', 'The breakthrough in the 1980s showed that adding a second layer of neurons made it mathematically provable that any mathematical model could be approximated to any level of accuracy with neural networks, signifying a significant advancement in the capabilities of neural networks. Mathematically proven capability to approximate any mathematical model to any level of accuracy with neural networks.', 'The course dispels the myth that deep learning requires extensive resources and technical expertise, highlighting that world-class research and industry projects have been achieved using minimal resources and expertise.', 'Neural networks have been widely used in industry since the early 90s, particularly in big companies for tasks such as targeted marketing for retail banks, demonstrating their early adoption and impact in various industries. Widespread adoption in industry, particularly in big companies for various applications.']}, {'end': 1848.42, 'segs': [{'end': 962.739, 'src': 'embed', 'start': 920.785, 'weight': 0, 'content': [{'end': 925.647, 'text': 'So when you add more layers to a neural network, you get deep learning.', 'start': 920.785, 'duration': 4.862}, {'end': 931.289, 'text': "So deep doesn't mean anything like mystical, it just means more layers.", 'start': 926.027, 'duration': 5.262}, {'end': 933.99, 'text': 'More layers than just adding the one extra one.', 'start': 932.029, 'duration': 1.961}, {'end': 942.226, 'text': "So thanks to that, neural nets are now living up to their potential, as we saw in that, like, what's deep learning good at thing.", 'start': 935.723, 'duration': 6.503}, {'end': 945.308, 'text': 'So we could now say that Rosenblatt was right.', 'start': 942.606, 'duration': 2.702}, {'end': 954.896, 'text': "We have a machine that's capable of perceiving recognizing and identifying its surroundings without any human training or control.", 'start': 945.988, 'duration': 8.908}, {'end': 956.537, 'text': "That's definitely true.", 'start': 955.756, 'duration': 0.781}, {'end': 960.218, 'text': "I don't think there's anything controversial about that statement based on the current technology.", 'start': 956.577, 'duration': 3.641}, {'end': 962.739, 'text': "So we're going to be learning how to do that.", 'start': 961.399, 'duration': 1.34}], 'summary': 'Adding more layers to a neural network leads to deep learning, enabling machines to perceive, recognize, and identify surroundings without human control.', 'duration': 41.954, 'max_score': 920.785, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/_QUEXsHfsA0/pics/_QUEXsHfsA0920785.jpg'}, {'end': 1020.902, 'src': 'embed', 'start': 989.487, 'weight': 2, 'content': [{'end': 1000.517, 'text': 'And the reason for that is that people who study how to teach and learn have found that is not the right way to do it for most people.', 'start': 989.487, 'duration': 11.03}, {'end': 1004.449, 'text': 'For most people.', 'start': 1001.567, 'duration': 2.882}, {'end': 1013.536, 'text': 'so we work a lot based on the work of Professor David Perkins from Harvard and others who work at similar things,', 'start': 1004.449, 'duration': 9.087}, {'end': 1016.799, 'text': 'who talk about this idea of playing the whole game.', 'start': 1013.536, 'duration': 3.263}, {'end': 1020.902, 'text': "And so playing the whole game is like, it's based on the sports analogy.", 'start': 1017.459, 'duration': 3.443}], 'summary': 'Teaching and learning should focus on playing the whole game, according to research by professor david perkins from harvard.', 'duration': 31.415, 'max_score': 989.487, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/_QUEXsHfsA0/pics/_QUEXsHfsA0989487.jpg'}, {'end': 1178.02, 'src': 'embed', 'start': 1148.583, 'weight': 3, 'content': [{'end': 1154.408, 'text': "So this is all about making sure that the thing you're doing, you're doing it properly.", 'start': 1148.583, 'duration': 5.825}, {'end': 1158.751, 'text': "You know, you're making it the whole thing.", 'start': 1154.948, 'duration': 3.803}, {'end': 1160.573, 'text': "You're providing the context and the interest.", 'start': 1158.771, 'duration': 1.802}, {'end': 1172.839, 'text': "Um, so for the fast AI approach to, to learning deep learning, what this means is that today we're going to train models end to end.", 'start': 1162.316, 'duration': 10.523}, {'end': 1178.02, 'text': "We're going to actually train models, right? And they won't just be crappy models.", 'start': 1173.479, 'duration': 4.541}], 'summary': 'Ensuring proper execution and understanding, training end-to-end models for deep learning with fast ai approach.', 'duration': 29.437, 'max_score': 1148.583, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/_QUEXsHfsA0/pics/_QUEXsHfsA01148583.jpg'}, {'end': 1239.897, 'src': 'embed', 'start': 1207.868, 'weight': 4, 'content': [{'end': 1209.189, 'text': 'Yeah, deliberate practice right?', 'start': 1207.868, 'duration': 1.321}, {'end': 1215.792, 'text': "Work on the hard parts means that You don't just.", 'start': 1209.91, 'duration': 5.882}, {'end': 1219.332, 'text': 'you know, swing a bat, you know at a ball every time.', 'start': 1215.792, 'duration': 3.54}, {'end': 1223.873, 'text': 'you know you go out and just muck around, but you train properly.', 'start': 1219.332, 'duration': 4.541}, {'end': 1227.034, 'text': "You find the bit that you're the least good at you.", 'start': 1223.893, 'duration': 3.141}, {'end': 1228.574, 'text': 'figure out where the problems are.', 'start': 1227.034, 'duration': 1.54}, {'end': 1230.395, 'text': 'you work damn hard at it, right?', 'start': 1228.574, 'duration': 1.821}, {'end': 1239.897, 'text': 'So, in the deep learning context, that means that we do not dumb things down right?', 'start': 1230.855, 'duration': 9.042}], 'summary': 'Deliberate practice involves identifying and working on the least proficient areas to improve skills, not dumbing things down.', 'duration': 32.029, 'max_score': 1207.868, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/_QUEXsHfsA0/pics/_QUEXsHfsA01207868.jpg'}, {'end': 1464.898, 'src': 'embed', 'start': 1433.89, 'weight': 5, 'content': [{'end': 1437.292, 'text': 'It got super bogged down.', 'start': 1433.89, 'duration': 3.402}, {'end': 1439.644, 'text': 'basically TensorFlow got super bogged down.', 'start': 1437.642, 'duration': 2.002}, {'end': 1447.171, 'text': 'This other software called PyTorch came along that was much easier to use and much more useful to researchers.', 'start': 1439.664, 'duration': 7.507}, {'end': 1460.696, 'text': 'And within the last 12 months, the the percentage of papers at major conferences that uses PyTorch has gone from 20% to 80% and vice versa.', 'start': 1448.592, 'duration': 12.104}, {'end': 1464.898, 'text': 'Those that use TensorFlow have gone from 80% to 20%.', 'start': 1460.856, 'duration': 4.042}], 'summary': 'Pytorch usage in major conferences increased from 20% to 80% within 12 months, while tensorflow usage decreased from 80% to 20%.', 'duration': 31.008, 'max_score': 1433.89, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/_QUEXsHfsA0/pics/_QUEXsHfsA01433890.jpg'}, {'end': 1520.822, 'src': 'embed', 'start': 1487.015, 'weight': 6, 'content': [{'end': 1489.737, 'text': "It's certainly not designed for beginner friendliness.", 'start': 1487.015, 'duration': 2.722}, {'end': 1496.902, 'text': "And it's not designed for what we would say it doesn't have, like higher level APIs,", 'start': 1492.419, 'duration': 4.483}, {'end': 1505.299, 'text': "by which I mean there isn't really things to make it easy to build stuff quickly using PyTorch.", 'start': 1496.902, 'duration': 8.397}, {'end': 1512.58, 'text': 'So to deal with that issue, we have a library called Fast.ai that sits on top of PyTorch.', 'start': 1505.699, 'duration': 6.881}, {'end': 1520.822, 'text': 'Fast.ai is the most popular higher level API for PyTorch.', 'start': 1513.36, 'duration': 7.462}], 'summary': 'Pytorch lacks beginner friendliness and higher level apis, addressed by fast.ai.', 'duration': 33.807, 'max_score': 1487.015, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/_QUEXsHfsA0/pics/_QUEXsHfsA01487015.jpg'}, {'end': 1711.564, 'src': 'embed', 'start': 1685.279, 'weight': 7, 'content': [{'end': 1695.809, 'text': "It's just a huge distraction to be spending your time doing system administration on a GPU machine and installing drivers and blah blah blah.", 'start': 1685.279, 'duration': 10.53}, {'end': 1698.311, 'text': 'And run it on Linux, please.', 'start': 1696.269, 'duration': 2.042}, {'end': 1700.633, 'text': "That's what everybody's doing, not just us.", 'start': 1698.831, 'duration': 1.802}, {'end': 1702.134, 'text': "Everybody's running it on Linux.", 'start': 1700.693, 'duration': 1.441}, {'end': 1703.255, 'text': 'Make life easy for yourself.', 'start': 1702.154, 'duration': 1.101}, {'end': 1704.737, 'text': "It's hard enough to learn deep learning.", 'start': 1703.315, 'duration': 1.422}, {'end': 1711.564, 'text': "without having to do it in a way that you're learning all kinds of arcane hardware support issues.", 'start': 1705.157, 'duration': 6.407}], 'summary': 'Running deep learning on linux makes life easier, avoiding hardware support issues.', 'duration': 26.285, 'max_score': 1685.279, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/_QUEXsHfsA0/pics/_QUEXsHfsA01685279.jpg'}, {'end': 1793.528, 'src': 'embed', 'start': 1760.921, 'weight': 8, 'content': [{'end': 1767.527, 'text': 'Most of the other systems we recommend save your work for you automatically, and you can come back to it any time.', 'start': 1760.921, 'duration': 6.606}, {'end': 1768.587, 'text': "Colab doesn't.", 'start': 1767.927, 'duration': 0.66}, {'end': 1776.353, 'text': 'So be sure to check out the Colab platform thread on the forums to learn about that.', 'start': 1769.068, 'duration': 7.285}, {'end': 1779.396, 'text': 'So I mentioned the forums.', 'start': 1777.634, 'duration': 1.762}, {'end': 1793.528, 'text': 'The forums are really, really important because that is where all of the discussion and setup and everything happens.', 'start': 1783.995, 'duration': 9.533}], 'summary': "Colab doesn't auto-save work. forums are vital for discussion and setup.", 'duration': 32.607, 'max_score': 1760.921, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/_QUEXsHfsA0/pics/_QUEXsHfsA01760921.jpg'}], 'start': 895.761, 'title': 'Deep learning fundamentals and tools', 'summary': 'Covers deep learning basics, emphasizing the importance of multiple layers in neural networks and a top-down teaching approach. it also introduces fast ai learning, promoting a non-traditional educational approach, and discusses python, tensorflow, pytorch, nvidia gpu, linux, and community forums in the deep learning context.', 'chapters': [{'end': 1147.322, 'start': 895.761, 'title': 'Deep learning basics and effective teaching', 'summary': 'Discusses the concept of deep learning, emphasizing the importance of multiple layers in neural networks, enabling the potential of neural nets, and advocating a top-down approach to teaching based on the work of professor david perkins from harvard.', 'duration': 251.561, 'highlights': ['Neural networks require multiple layers for practical good performance, as shown by researchers 30 years ago, contrary to the belief that one extra layer could suffice, leading to the emergence of deep learning.', 'The potential of neural nets is being realized, enabling capabilities such as perceiving, recognizing, and identifying surroundings without human training or control.', 'The chapter advocates a top-down approach to teaching, aligning with the work of Professor David Perkins from Harvard, emphasizing the effectiveness of playing the whole game and making the game worth playing in the learning process.']}, {'end': 1381.66, 'start': 1148.583, 'title': 'Fast ai learning approach', 'summary': 'Introduces the fast ai approach to learning deep learning, emphasizing the training of state-of-the-art models and deliberate practice on challenging aspects, promoting a non-traditional educational approach.', 'duration': 233.077, 'highlights': ['The fast AI approach emphasizes training state-of-the-art world class models, ensuring a practical and comprehensive learning experience.', 'Deliberate practice and working on the hard parts are highlighted, focusing on training properly and addressing the least proficient areas.', 'The non-traditional educational approach is encouraged, promoting discomfort with unfamiliarity and the use of software without comprehensive understanding, challenging but valuable for learning deep learning.', 'Teaching using the fast AI approach is acknowledged as very challenging but potentially worth it, indicating the difficulty in deviating from a traditional foundations-first approach.']}, {'end': 1848.42, 'start': 1381.72, 'title': 'Deep learning software and tools', 'summary': 'Discusses the dominance of python, the shift from tensorflow to pytorch, the importance of fast.ai, the flexibility of pytorch, the need for nvidia gpu, the preference for linux, and the significance of the forums in the deep learning community.', 'duration': 466.7, 'highlights': ['PyTorch usage has increased from 20% to 80% in the last 12 months, while TensorFlow usage has decreased from 80% to 20%. The rapid change in the usage of PyTorch and TensorFlow, with PyTorch being favored by the majority of deep learning practitioners and researchers, indicates its growing dominance in the field.', 'Fast.ai is the most popular higher level API for PyTorch, designed for beginners, practitioners, and researchers. The versatility of Fast.ai, catering to beginners, industry practitioners, and researchers, positions it as the leading higher-level API for PyTorch, reflecting its widespread usage and applicability.', 'The importance of using an NVIDIA GPU and running deep learning on Linux for better support and ease of use. The necessity of an NVIDIA GPU and Linux for deep learning is emphasized to ensure optimal performance and minimal hardware support issues, underlining the industry standard for deep learning setups.', 'The significance of the forums for discussions, setup help, and accessing step-by-step instructions for setting up servers and platforms. The forums serve as a vital platform for discussions, setup assistance, and comprehensive instructions, fostering collaboration and providing support for individuals navigating the deep learning landscape.']}], 'duration': 952.659, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/_QUEXsHfsA0/pics/_QUEXsHfsA0895761.jpg', 'highlights': ['Deep learning relies on multiple layers in neural networks for practical good performance, contrary to the belief that one extra layer could suffice, leading to the emergence of deep learning.', 'The potential of neural nets enables capabilities such as perceiving, recognizing, and identifying surroundings without human training or control.', 'The chapter advocates a top-down approach to teaching, aligning with the work of Professor David Perkins from Harvard, emphasizing the effectiveness of playing the whole game and making the game worth playing in the learning process.', 'The fast AI approach emphasizes training state-of-the-art world class models, ensuring a practical and comprehensive learning experience.', 'Deliberate practice and working on the hard parts are highlighted, focusing on training properly and addressing the least proficient areas.', 'PyTorch usage has increased from 20% to 80% in the last 12 months, while TensorFlow usage has decreased from 80% to 20%, indicating its growing dominance in the field.', 'Fast.ai is the most popular higher level API for PyTorch, designed for beginners, practitioners, and researchers, reflecting its widespread usage and applicability.', 'The importance of using an NVIDIA GPU and running deep learning on Linux for better support and ease of use, underlining the industry standard for deep learning setups.', 'The significance of the forums for discussions, setup help, and accessing step-by-step instructions for setting up servers and platforms, fostering collaboration and providing support for individuals navigating the deep learning landscape.']}, {'end': 2723.757, 'segs': [{'end': 1972.618, 'src': 'embed', 'start': 1938.964, 'weight': 0, 'content': [{'end': 1953.757, 'text': 'The Jupyter Notebook REPL is particularly interesting because it has things like, headings, graphical outputs, interactive, multimedia.', 'start': 1938.964, 'duration': 14.793}, {'end': 1956.54, 'text': "It's a really astonishing piece of software.", 'start': 1954.057, 'duration': 2.483}, {'end': 1959.182, 'text': "It's won some really big awards.", 'start': 1957.32, 'duration': 1.862}, {'end': 1967.671, 'text': "It's, you know, I would have thought the most widely used REPL outside of shells like Bash.", 'start': 1959.443, 'duration': 8.228}, {'end': 1969.835, 'text': "It's a very powerful system.", 'start': 1968.814, 'duration': 1.021}, {'end': 1970.416, 'text': 'We love it.', 'start': 1969.975, 'duration': 0.441}, {'end': 1972.618, 'text': "We've written our whole book in it.", 'start': 1971.176, 'duration': 1.442}], 'summary': 'Jupyter notebook: award-winning, widely used repl with powerful features and multimedia capabilities.', 'duration': 33.654, 'max_score': 1938.964, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/_QUEXsHfsA0/pics/_QUEXsHfsA01938964.jpg'}, {'end': 2424.074, 'src': 'embed', 'start': 2401.78, 'weight': 2, 'content': [{'end': 2412.306, 'text': 'So the idea with this is this is probably the version that you want to be experimenting with, because it kind of forces you to think about,', 'start': 2401.78, 'duration': 10.526}, {'end': 2418.63, 'text': "like what's going on, as you do each step you know, rather than just reading it and running it without thinking.", 'start': 2412.306, 'duration': 6.324}, {'end': 2424.074, 'text': "We kind of want you to to do it in this more bare environment in which you're thinking about, like oh,", 'start': 2418.73, 'duration': 5.344}], 'summary': 'Encourages experimentation and critical thinking in a more deliberate learning environment.', 'duration': 22.294, 'max_score': 2401.78, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/_QUEXsHfsA0/pics/_QUEXsHfsA02401780.jpg'}, {'end': 2568.03, 'src': 'embed', 'start': 2540.103, 'weight': 3, 'content': [{'end': 2544.564, 'text': "All right, so let's try running the first part of the notebook.", 'start': 2540.103, 'duration': 4.461}, {'end': 2550.686, 'text': 'So here we are in 01 intro.', 'start': 2546.405, 'duration': 4.281}, {'end': 2558.247, 'text': 'So this is chapter 1 and here is our first cell.', 'start': 2551.686, 'duration': 6.561}, {'end': 2565.309, 'text': 'So I click on the cell and by default actually there will be a header and a toolbar.', 'start': 2558.267, 'duration': 7.042}, {'end': 2566.55, 'text': 'As you can see you can turn them on and off.', 'start': 2565.329, 'duration': 1.221}, {'end': 2568.03, 'text': 'I always leave them off myself.', 'start': 2566.89, 'duration': 1.14}], 'summary': 'Running the first part of the notebook in chapter 1 with header and toolbar options available.', 'duration': 27.927, 'max_score': 2540.103, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/_QUEXsHfsA0/pics/_QUEXsHfsA02540103.jpg'}, {'end': 2676.819, 'src': 'embed', 'start': 2645.162, 'weight': 4, 'content': [{'end': 2646.242, 'text': "What is the something it's doing??", 'start': 2645.162, 'duration': 1.08}, {'end': 2657.868, 'text': "Well, what it's doing here is it's actually grabbing a data set we call the pets data set, which is a data set of pictures of cats and dogs,", 'start': 2646.682, 'duration': 11.186}, {'end': 2664.992, 'text': "And it's trying to figure out which ones are cats and which ones are dogs.", 'start': 2658.728, 'duration': 6.264}, {'end': 2673.738, 'text': "And as you can see, after about, well, less than a minute, it's able to do that with a 0.5% error rate.", 'start': 2666.093, 'duration': 7.645}, {'end': 2676.819, 'text': 'So it can do it pretty much perfectly.', 'start': 2673.778, 'duration': 3.041}], 'summary': 'Using the pets dataset, it distinguishes cats and dogs with a 0.5% error rate.', 'duration': 31.657, 'max_score': 2645.162, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/_QUEXsHfsA0/pics/_QUEXsHfsA02645162.jpg'}], 'start': 1848.42, 'title': 'Jupyter notebook features and usage', 'summary': 'Introduces jupyter notebook, emphasizing its powerful repl system for python coding, graphical outputs, widespread usage for teaching and development. it also introduces the concept of jupyter notebook, explaining its repl-based system, modes, keyboard shortcuts, and integration with github. additionally, it highlights the importance of experimenting with content, use of questionnaires, and practical advice for running the notebook and training models.', 'chapters': [{'end': 1994.612, 'start': 1848.42, 'title': 'Introduction to jupyter notebook', 'summary': 'Introduces jupyter notebook, a powerful repl system for python coding, emphasizing its features like graphical outputs and its widespread usage for teaching and development.', 'duration': 146.192, 'highlights': ['Jupyter Notebook allows you to type Python code and obtain results by pressing shift enter, making it a powerful tool for coding and interactive learning.', 'The REPL system in Jupyter Notebook offers features such as headings, graphical outputs, and multimedia, making it a widely used and powerful system for teaching and development.', "The chapter emphasizes Jupyter Notebook's significance by highlighting its use in writing the entire fast AI library and teaching, indicating its widespread adoption and influence in the coding community."]}, {'end': 2374.211, 'start': 1994.632, 'title': 'Understanding jupyter notebook', 'summary': 'Introduces the concept of jupyter notebook, explaining its repl-based system, two modes (edit and command), keyboard shortcuts, and its usefulness in creating formatted text, plots, and lists. it also mentions its integration with github and the importance of creating duplicates to avoid conflicts with updates.', 'duration': 379.579, 'highlights': ['The chapter introduces the concept of Jupyter Notebook, explaining its REPL-based system, two modes (edit and command), keyboard shortcuts, and its usefulness in creating formatted text, plots, and lists. It also mentions its integration with GitHub and the importance of creating duplicates to avoid conflicts with updates.', 'It explains the concept of REPL-based systems and emphasizes the usefulness of Jupyter Notebook in creating formatted text, plots, lists, and integrating with GitHub.', 'It details the two modes of Jupyter Notebook, edit and command, and highlights the keyboard shortcuts available in command mode, such as copying, pasting, cutting, adding new cells, creating headings, and typing formatted text in Markdown.', 'The chapter underscores the importance of creating duplicates of notebooks to avoid conflicts with updates and mentions the integration of Jupyter Notebook with GitHub for updating course materials.', 'It provides practical advice on creating duplicates of notebooks to avoid conflicts with updates and emphasizes the integration of Jupyter Notebook with GitHub for updating course materials.']}, {'end': 2723.757, 'start': 2374.612, 'title': 'Fastbook chapter 1 summary', 'summary': 'Highlights the importance of experimenting with the content, emphasizes the use of questionnaires to grasp key takeaways, and provides practical advice for running the notebook and training models, including potential troubleshooting steps.', 'duration': 349.145, 'highlights': ['The chapter emphasizes the importance of experimenting with the content and thinking through each step, rather than just reading and running it, to facilitate deeper learning.', 'The use of questionnaires is highlighted as a tool to understand the important takeaways from each notebook, encouraging readers to complete them before moving on to the next chapter.', 'Practical advice for running the notebook is provided, including guidance on running cells, managing headers and toolbars, and interpreting progress bars.', 'The process of training a model is explained, with practical advice to not expect identical results due to randomness, potential troubleshooting steps, and the demonstration of training a model to classify pictures of cats and dogs with a low error rate.', 'Readers are advised on the use of GPUs, with specific mention of potential issues with Windows and Apple hardware, and the recommendation to use Linux for easier functionality.']}], 'duration': 875.337, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/_QUEXsHfsA0/pics/_QUEXsHfsA01848420.jpg', 'highlights': ["Jupyter Notebook's REPL system offers features such as headings, graphical outputs, and multimedia, making it a widely used and powerful system for teaching and development.", 'The chapter introduces the concept of Jupyter Notebook, explaining its REPL-based system, two modes (edit and command), keyboard shortcuts, and its usefulness in creating formatted text, plots, and lists.', 'The chapter emphasizes the importance of experimenting with the content and thinking through each step, rather than just reading and running it, to facilitate deeper learning.', 'Practical advice for running the notebook is provided, including guidance on running cells, managing headers and toolbars, and interpreting progress bars.', 'The process of training a model is explained, with practical advice to not expect identical results due to randomness, potential troubleshooting steps, and the demonstration of training a model to classify pictures of cats and dogs with a low error rate.']}, {'end': 3412.886, 'segs': [{'end': 2791.226, 'src': 'embed', 'start': 2752.572, 'weight': 0, 'content': [{'end': 2754.833, 'text': 'We can even create GUIs in this REPL.', 'start': 2752.572, 'duration': 2.261}, {'end': 2759.352, 'text': 'So if I click on this file upload, And I can pick cat.', 'start': 2755.233, 'duration': 4.119}, {'end': 2762.493, 'text': 'There we go.', 'start': 2762.053, 'duration': 0.44}, {'end': 2769.698, 'text': 'And I can now turn that uploaded data into an image.', 'start': 2763.794, 'duration': 5.904}, {'end': 2773.4, 'text': "There's a cat.", 'start': 2772.759, 'duration': 0.641}, {'end': 2775.681, 'text': 'And now I can do predict.', 'start': 2774.24, 'duration': 1.441}, {'end': 2780.764, 'text': "And it's a cat with a 99.96% probability.", 'start': 2778.263, 'duration': 2.501}, {'end': 2789.765, 'text': "So we can see we have just uploaded an image that we've picked out.", 'start': 2786.323, 'duration': 3.442}, {'end': 2791.226, 'text': 'So you should try this right?', 'start': 2789.825, 'duration': 1.401}], 'summary': 'Using repl, the speaker demonstrates uploading an image, which is identified as a cat with 99.96% probability.', 'duration': 38.654, 'max_score': 2752.572, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/_QUEXsHfsA0/pics/_QUEXsHfsA02752572.jpg'}, {'end': 3412.886, 'src': 'embed', 'start': 3359.739, 'weight': 1, 'content': [{'end': 3369.661, 'text': 'We want a completely general way to do this to update the weights based on some measure of performance,', 'start': 3359.739, 'duration': 9.922}, {'end': 3372.842, 'text': 'such as how good is it at recognizing cats versus dogs?', 'start': 3369.661, 'duration': 3.181}, {'end': 3376.542, 'text': 'And luckily it turns out such a thing exists.', 'start': 3374.102, 'duration': 2.44}, {'end': 3381.363, 'text': 'And that thing is called Stochastic Gradient Descent or SGD.', 'start': 3377.762, 'duration': 3.601}, {'end': 3383.784, 'text': "Again, we'll look at exactly how it works.", 'start': 3381.743, 'duration': 2.041}, {'end': 3385.864, 'text': "We'll build it from ourselves from scratch.", 'start': 3384.184, 'duration': 1.68}, {'end': 3388.122, 'text': "for now we don't have to worry about it.", 'start': 3386.602, 'duration': 1.52}, {'end': 3394.863, 'text': 'I will tell you this though, neither SGD nor neural nets are at all mathematically complex.', 'start': 3388.142, 'duration': 6.721}, {'end': 3398.784, 'text': 'They nearly entirely are addition and multiplication.', 'start': 3395.344, 'duration': 3.44}, {'end': 3406.305, 'text': 'The trick is it just does a lot of them, like billions of them, so many more than we can like intuitively grasp.', 'start': 3399.504, 'duration': 6.801}, {'end': 3412.886, 'text': "They can do extraordinarily powerful things but they're not, they're not rocket science at all.", 'start': 3406.325, 'duration': 6.561}], 'summary': 'Stochastic gradient descent updates weights based on performance, not mathematically complex, involves billions of additions and multiplications.', 'duration': 53.147, 'max_score': 3359.739, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/_QUEXsHfsA0/pics/_QUEXsHfsA03359739.jpg'}], 'start': 2727.199, 'title': 'Machine learning and image recognition', 'summary': 'Covers a demonstration of an image recognition model achieving 99.96% accuracy in distinguishing cats from dogs and an introduction to machine learning, including its historical development, training models using weights, the role of neural networks, and the process of training models using stochastic gradient descent.', 'chapters': [{'end': 2841.1, 'start': 2727.199, 'title': 'Image recognition model demonstration', 'summary': 'Demonstrates training a model to recognize cats from dogs, achieving 99.96% accuracy, and creating a gui for image uploading and prediction.', 'duration': 113.901, 'highlights': ["We've trained a model to pick cats from dogs, achieving 99.96% accuracy in recognizing a cat from an uploaded image.", 'The demonstration includes creating a GUI in the REPL for image uploading and prediction, showcasing the interactive capabilities of the environment.', "The course emphasizes the model's ability to learn from specific types of information, such as photos of real cats, and not from other representations like anime cats or drawings."]}, {'end': 3412.886, 'start': 2846.144, 'title': 'Introduction to machine learning', 'summary': 'Introduces the concept of machine learning, starting from the challenges of traditional programming to the development of machine learning by arthur samuel in 1962, leading to the concept of training models using weights and the role of neural networks in solving any solvable problem. it also discusses the process of training models using stochastic gradient descent, emphasizing the simplicity of neural networks and their potential power.', 'duration': 566.742, 'highlights': ["The concept of machine learning developed by Arthur Samuel in 1962, focusing on training models using weights instead of explicitly programming steps, leading to the use of neural networks to solve any solvable problem. Arthur Samuel's approach in 1962 shifted the focus from traditional programming to training models using weights, leading to the development of neural networks capable of solving any solvable problem with the right set of weights.", 'The introduction of Stochastic Gradient Descent as a method to update weights based on performance, providing a general way to train models in machine learning. Stochastic Gradient Descent (SGD) is introduced as a method to update weights based on performance, offering a general approach to training models in machine learning.', 'The simplicity of neural networks, primarily involving addition and multiplication, but capable of performing billions of operations to achieve powerful results. Neural networks are described as simple in terms of their mathematical complexity, primarily involving addition and multiplication, yet capable of performing billions of operations, showcasing their potential power.']}], 'duration': 685.687, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/_QUEXsHfsA0/pics/_QUEXsHfsA02727199.jpg', 'highlights': ['Trained model achieves 99.96% accuracy in recognizing cats from dogs.', 'Introduction of Stochastic Gradient Descent as a method to update weights based on performance.', 'Neural networks are described as simple in terms of their mathematical complexity, yet capable of performing billions of operations.']}, {'end': 3688.197, 'segs': [{'end': 3443.242, 'src': 'embed', 'start': 3413.107, 'weight': 0, 'content': [{'end': 3416.027, 'text': "They're not complex things and we'll see exactly how they work.", 'start': 3413.107, 'duration': 2.92}, {'end': 3422.625, 'text': "Right, so that's the Arthur Samuel version, right.", 'start': 3418.123, 'duration': 4.502}, {'end': 3427.266, 'text': "Nowadays we don't use quite the same terminology, but we use exactly the same idea.", 'start': 3422.985, 'duration': 4.281}, {'end': 3434.309, 'text': 'So that function that sits in the middle we call an architecture.', 'start': 3428.587, 'duration': 5.722}, {'end': 3439.991, 'text': "An architecture is the function that we're adjusting the weights to get it to do something.", 'start': 3434.489, 'duration': 5.502}, {'end': 3443.242, 'text': "So that's the architecture, that's the functional form of the model.", 'start': 3440.56, 'duration': 2.682}], 'summary': 'Exploring the architecture and functional form of models.', 'duration': 30.135, 'max_score': 3413.107, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/_QUEXsHfsA0/pics/_QUEXsHfsA03413107.jpg'}, {'end': 3560.918, 'src': 'embed', 'start': 3490.624, 'weight': 1, 'content': [{'end': 3491.825, 'text': 'So the results are predictions.', 'start': 3490.624, 'duration': 1.201}, {'end': 3498.439, 'text': "the measure of performance, to use Arthur Samuel's word, is known as the loss.", 'start': 3492.875, 'duration': 5.564}, {'end': 3502.021, 'text': 'So the loss has been calculated from the labels and the predictions.', 'start': 3499.219, 'duration': 2.802}, {'end': 3506.223, 'text': "Okay, and then there's the update back to the parameters.", 'start': 3503.001, 'duration': 3.222}, {'end': 3513.408, 'text': 'Okay, so this is the same picture as we saw, but just putting in the words that we use today.', 'start': 3508.104, 'duration': 5.304}, {'end': 3522.144, 'text': 'So this picture, if you forget, if I say these are the parameters of this you know, used for this architecture to create a model,', 'start': 3515.009, 'duration': 7.135}, {'end': 3527.208, 'text': 'you can go back and remind yourself what did I mean? What are the parameters? What are the predictions? What is the loss?', 'start': 3522.144, 'duration': 5.064}, {'end': 3535.114, 'text': 'Okay, the loss is some function that measures the performance of the model in such a way that we can update the parameters.', 'start': 3527.228, 'duration': 7.886}, {'end': 3545.015, 'text': "It's important to note that deep learning and machine learning are not magic.", 'start': 3540.734, 'duration': 4.281}, {'end': 3554.377, 'text': "The model can only be created where you have data showing you examples of the thing that you're trying to learn about.", 'start': 3546.195, 'duration': 8.182}, {'end': 3560.918, 'text': "It can only learn to operate on the patterns that you've seen in the input used to train it.", 'start': 3554.397, 'duration': 6.521}], 'summary': 'Loss measures model performance; data essential for deep learning.', 'duration': 70.294, 'max_score': 3490.624, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/_QUEXsHfsA0/pics/_QUEXsHfsA03490624.jpg'}, {'end': 3666.97, 'src': 'embed', 'start': 3601.857, 'weight': 3, 'content': [{'end': 3604.617, 'text': "Well, I don't know, we don't do that right?", 'start': 3601.857, 'duration': 2.76}, {'end': 3612.039, 'text': "We can predict what somebody will say about a product after we've shown them, but we're not creating actions, we're creating predictions.", 'start': 3604.897, 'duration': 7.142}, {'end': 3615.5, 'text': "That's a super important difference to recognize.", 'start': 3612.159, 'duration': 3.341}, {'end': 3624.829, 'text': "It's not enough just to have examples of input data, like pictures of dogs and cats, We can't do anything without labels.", 'start': 3617.58, 'duration': 7.249}, {'end': 3630.136, 'text': "And so very often organizations say we don't have enough data.", 'start': 3626.211, 'duration': 3.925}, {'end': 3634.181, 'text': "Most of the time they mean we don't have enough labeled data.", 'start': 3631.077, 'duration': 3.104}, {'end': 3638.712, 'text': 'because if a company is trying to do something with deep learning,', 'start': 3634.949, 'duration': 3.763}, {'end': 3644.697, 'text': "often it's because they're trying to automate or improve something they're already doing, which means, by definition,", 'start': 3638.712, 'duration': 5.985}, {'end': 3650.581, 'text': "they have data about that thing or a way to capture data about that thing, because they're doing it right?", 'start': 3644.697, 'duration': 5.884}, {'end': 3654.805, 'text': 'But often the tricky part is labeling it.', 'start': 3651.342, 'duration': 3.463}, {'end': 3659.849, 'text': "So, for example, in medicine, if you're trying to build a model for radiology,", 'start': 3655.025, 'duration': 4.824}, {'end': 3666.97, 'text': 'you can almost certainly get lots of medical images about just about anything you can think of,', 'start': 3660.728, 'duration': 6.242}], 'summary': 'Creating predictions requires labeled data, often lacking in organizations.', 'duration': 65.113, 'max_score': 3601.857, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/_QUEXsHfsA0/pics/_QUEXsHfsA03601857.jpg'}], 'start': 3413.107, 'title': 'Neural network architecture basics and challenges in deep learning', 'summary': 'Explains fundamental concepts of neural network architecture, including terminology, functions, and key components, and discusses the importance of labeled data in deep learning, emphasizing the learning approach and difficulty in obtaining labeled data, with a focus on the medical field.', 'chapters': [{'end': 3535.114, 'start': 3413.107, 'title': 'Neural network architecture basics', 'summary': 'Explains the fundamental concepts of neural network architecture, including terminology, functions, and key components such as parameters, predictions, and loss.', 'duration': 122.007, 'highlights': ['The function that sits in the middle of the architecture is what we call an architecture, which is adjusted by the parameters to produce predictions.', 'The predictions, based on independent and dependent variables, are measured by the loss function, which is used to update the parameters.', 'The terminology in neural network architecture includes parameters (weights), predictions (based on input variables), and loss (measure of model performance).']}, {'end': 3688.197, 'start': 3540.734, 'title': 'Challenges in deep learning', 'summary': 'Discusses the importance of labeled data in deep learning, emphasizing that the learning approach only creates predictions and not actions, and the difficulty in obtaining labeled data, with a focus on the medical field.', 'duration': 147.463, 'highlights': ['Lack of labeled data is a common challenge in deep learning, as organizations often struggle to obtain enough labeled data for training models.', 'Deep learning models can only learn from examples of the thing being learned about, and the lack of labeled data prevents the model from making updates to its parameters, hindering its ability to improve predictions.', 'The learning approach in deep learning only results in creating predictions, not actions, which is crucial to recognize, particularly in fields such as medical imaging where labeling data is a complex task.', 'Obtaining labeled data in fields like medicine, especially in radiology, can be difficult due to the lack of structured capturing of labels, impacting the strategy for developing deep learning models.']}], 'duration': 275.09, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/_QUEXsHfsA0/pics/_QUEXsHfsA03413107.jpg', 'highlights': ['The function that sits in the middle of the architecture is what we call an architecture, which is adjusted by the parameters to produce predictions.', 'The predictions, based on independent and dependent variables, are measured by the loss function, which is used to update the parameters.', 'The terminology in neural network architecture includes parameters (weights), predictions (based on input variables), and loss (measure of model performance).', 'Lack of labeled data is a common challenge in deep learning, as organizations often struggle to obtain enough labeled data for training models.', 'Deep learning models can only learn from examples of the thing being learned about, and the lack of labeled data prevents the model from making updates to its parameters, hindering its ability to improve predictions.', 'The learning approach in deep learning only results in creating predictions, not actions, which is crucial to recognize, particularly in fields such as medical imaging where labeling data is a complex task.', 'Obtaining labeled data in fields like medicine, especially in radiology, can be difficult due to the lack of structured capturing of labels, impacting the strategy for developing deep learning models.']}, {'end': 4444.767, 'segs': [{'end': 3764.826, 'src': 'embed', 'start': 3742.954, 'weight': 0, 'content': [{'end': 3751.279, 'text': 'black people in the US, I think, get arrested something like seven times more often for, say, marijuana possession than whites,', 'start': 3742.954, 'duration': 8.325}, {'end': 3757.462, 'text': 'even though the actual underlying amount of marijuana use is about the same in the two populations.', 'start': 3751.279, 'duration': 6.183}, {'end': 3764.826, 'text': 'So if you start with biased data and you build a predictive policing model, its prediction will say oh,', 'start': 3757.842, 'duration': 6.984}], 'summary': 'Black people in the us are arrested 7 times more for marijuana possession than whites, despite similar usage rates.', 'duration': 21.872, 'max_score': 3742.954, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/_QUEXsHfsA0/pics/_QUEXsHfsA03742954.jpg'}, {'end': 3830.332, 'src': 'embed', 'start': 3785.339, 'weight': 1, 'content': [{'end': 3792.162, 'text': "will now find oh there's even more people we should be arresting in the black neighborhoods and thus it continues.", 'start': 3785.339, 'duration': 6.823}, {'end': 3799.466, 'text': 'So this would be an example of a modeling interacting with its environment to create something called a positive feedback loop,', 'start': 3792.562, 'duration': 6.904}, {'end': 3805.388, 'text': 'where the more a model is used, the more biased the data becomes, making the model even more biased, and so forth.', 'start': 3799.466, 'duration': 5.922}, {'end': 3818.061, 'text': 'So One of the things to be super careful about with machine learning is recognizing how that model is actually being used and what kinds of things might happen as a result of that.', 'start': 3806.509, 'duration': 11.552}, {'end': 3830.332, 'text': 'I was just going to add that this is also an example of proxies because here arrest is being used as a proxy for crime.', 'start': 3822.505, 'duration': 7.827}], 'summary': 'Using biased data in machine learning can create a positive feedback loop, leading to more biased models and potential misuse.', 'duration': 44.993, 'max_score': 3785.339, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/_QUEXsHfsA0/pics/_QUEXsHfsA03785339.jpg'}, {'end': 3996.404, 'src': 'embed', 'start': 3939.519, 'weight': 3, 'content': [{'end': 3945.261, 'text': 'we actually spent a lot of time figuring out how to avoid that problem so that you can import star safely.', 'start': 3939.519, 'duration': 5.742}, {'end': 3955.343, 'text': 'So whether you do this or not is entirely up to you, but rest assured that if you import star from a fastai library,', 'start': 3946.261, 'duration': 9.082}, {'end': 3960.851, 'text': "it's actually been explicitly designed in a way that you only get the bits that you actually need.", 'start': 3955.343, 'duration': 5.508}, {'end': 3965.652, 'text': "One thing to mention is in the video you see it's called FastAI 2.", 'start': 3961.952, 'duration': 3.7}, {'end': 3969.393, 'text': "That's because we're recording this video using a pre-release version.", 'start': 3965.652, 'duration': 3.741}, {'end': 3977.575, 'text': "By the time you're watching the online, the MOOC version of this, you will have, the 2 will be gone.", 'start': 3969.914, 'duration': 7.661}, {'end': 3989.681, 'text': 'Something else to mention is There are, as I speak, four main predefined applications in fastai, being vision, text,', 'start': 3978.856, 'duration': 10.825}, {'end': 3991.162, 'text': 'tabular and collaborative filtering.', 'start': 3989.681, 'duration': 1.481}, {'end': 3993.262, 'text': "We'll be learning about all of them and a lot more.", 'start': 3991.282, 'duration': 1.98}, {'end': 3996.404, 'text': 'For each one.', 'start': 3995.123, 'duration': 1.281}], 'summary': 'Fastai library designed to import star safely, with four main predefined applications: vision, text, tabular, and collaborative filtering.', 'duration': 56.885, 'max_score': 3939.519, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/_QUEXsHfsA0/pics/_QUEXsHfsA03939519.jpg'}, {'end': 4190.551, 'src': 'embed', 'start': 4165.617, 'weight': 5, 'content': [{'end': 4173.783, 'text': 'And as you can see, fastai already has predefined access to a number of really useful datasets, such as this pets dataset.', 'start': 4165.617, 'duration': 8.166}, {'end': 4179.627, 'text': 'Datasets are a super important part, as you can imagine, of deep learning.', 'start': 4175.725, 'duration': 3.902}, {'end': 4181.99, 'text': "We'll be seeing lots of them,", 'start': 4180.629, 'duration': 1.361}, {'end': 4190.551, 'text': 'and these are created by lots of heroes who basically spend months or years collating data that we can use to build these models.', 'start': 4181.99, 'duration': 8.561}], 'summary': 'Fastai has predefined access to numerous useful datasets, including the pets dataset, crucial for deep learning.', 'duration': 24.934, 'max_score': 4165.617, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/_QUEXsHfsA0/pics/_QUEXsHfsA04165617.jpg'}], 'start': 3691.761, 'title': 'Impact of biased data on predictive policing and importing libraries in python', 'summary': 'Discusses the impact of biased data on predictive policing models, emphasizing the sevenfold arrest rate disparity for marijuana possession among black people in the us, and also explains the significance of importing libraries in python, covering potential risks, fastai design, and key steps in using fastai for deep learning.', 'chapters': [{'end': 3862.931, 'start': 3691.761, 'title': 'Impact of biased data on predictive policing', 'summary': 'Discusses the impact of biased data on predictive policing models, highlighting that black people in the us are arrested seven times more often for marijuana possession than whites, despite similar marijuana use, leading to a positive feedback loop of biased data and biased predictions.', 'duration': 171.17, 'highlights': ['Black people in the US are arrested seven times more often than whites for marijuana possession, despite similar marijuana use, leading to biased data and predictions.', 'The use of biased data in predictive policing models creates a positive feedback loop, making the model even more biased as it is used.', 'The concept of proxies is discussed, where arrest is used as a proxy for crime, highlighting the significance of the difference between the proxy and the actual value in the data.']}, {'end': 4444.767, 'start': 3864.221, 'title': 'Importing libraries in python', 'summary': "Explains the importance of importing libraries in python, the potential risks of using 'import star', the design of fastai for rapid prototyping, and the key steps in using fastai for deep learning, including accessing datasets, creating a learner, and preventing overfitting through validation sets.", 'duration': 580.546, 'highlights': ["The design of fastai enables safe usage of 'import star' by importing only the necessary components, aligning with the practice of rapid prototyping and avoiding namespace issues (lines 11-21).", 'The chapter introduces the main predefined applications in fastai, including vision, text, tabular, and collaborative filtering, and emphasizes the availability of comprehensive documentation and tutorials for each application (lines 23-39).', 'The process of using fastai for deep learning involves accessing datasets, defining label functions, creating a learner with a specific architecture, and using validation sets to prevent overfitting (lines 83-97).']}], 'duration': 753.006, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/_QUEXsHfsA0/pics/_QUEXsHfsA03691761.jpg', 'highlights': ['Black people in the US are arrested seven times more often than whites for marijuana possession, despite similar marijuana use, leading to biased data and predictions.', 'The use of biased data in predictive policing models creates a positive feedback loop, making the model even more biased as it is used.', 'The concept of proxies is discussed, where arrest is used as a proxy for crime, highlighting the significance of the difference between the proxy and the actual value in the data.', "The design of fastai enables safe usage of 'import star' by importing only the necessary components, aligning with the practice of rapid prototyping and avoiding namespace issues (lines 11-21).", 'The chapter introduces the main predefined applications in fastai, including vision, text, tabular, and collaborative filtering, and emphasizes the availability of comprehensive documentation and tutorials for each application (lines 23-39).', 'The process of using fastai for deep learning involves accessing datasets, defining label functions, creating a learner with a specific architecture, and using validation sets to prevent overfitting (lines 83-97).']}, {'end': 4950.407, 'segs': [{'end': 4495.951, 'src': 'embed', 'start': 4445.507, 'weight': 3, 'content': [{'end': 4448.488, 'text': "So I've been programming, I mean since I was a kid, so like 40 years,", 'start': 4445.507, 'duration': 2.981}, {'end': 4459.758, 'text': 'and Sylvia and I both work really really hard to make Python do a lot of work for us and to use, you know,', 'start': 4450.829, 'duration': 8.929}, {'end': 4465.684, 'text': 'programming practices which make us very productive and allow us to come back to our code years later and still understand it.', 'start': 4459.758, 'duration': 5.926}, {'end': 4474.796, 'text': "And so you'll see, in our code we'll often do things that you might not have seen before,", 'start': 4467.046, 'duration': 7.75}, {'end': 4483.801, 'text': 'and so we a lot of students who have gone through previous courses say they learned a lot about coding and Python coding and software engineering from the course.', 'start': 4474.796, 'duration': 9.005}, {'end': 4486.343, 'text': 'So yeah, check, you know.', 'start': 4485.522, 'duration': 0.821}, {'end': 4491.366, 'text': "when you see something new, check it out and feel free to ask on the forums if you're curious about why something was done that way.", 'start': 4486.343, 'duration': 5.023}, {'end': 4495.951, 'text': 'One thing to mention is, Just like I mentioned,', 'start': 4493.347, 'duration': 2.604}], 'summary': 'Experienced programmers using python to boost productivity and learning outcomes for students.', 'duration': 50.444, 'max_score': 4445.507, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/_QUEXsHfsA0/pics/_QUEXsHfsA04445507.jpg'}, {'end': 4719.087, 'src': 'embed', 'start': 4636.341, 'weight': 0, 'content': [{'end': 4640.344, 'text': "so if you train this for a few minutes it's nearly perfect.", 'start': 4636.341, 'duration': 4.003}, {'end': 4650.369, 'text': 'But you can see, the basic idea is that we can very rapidly, with almost exactly the same code, create something not that classifies cats and dogs,', 'start': 4642.705, 'duration': 7.664}, {'end': 4654.352, 'text': "but does what's called segmentation, figures out what every pixel image is.", 'start': 4650.369, 'duration': 3.983}, {'end': 4661.673, 'text': "Look, here's the same thing, from import star, text loaded from folder, learner, learn fine tune.", 'start': 4655.731, 'duration': 5.942}, {'end': 4663.534, 'text': 'Same basic code.', 'start': 4662.614, 'duration': 0.92}, {'end': 4671.437, 'text': 'This is now something where we can give it a sentence and it can figure out whether that is expressing a positive or negative sentiment.', 'start': 4664.874, 'duration': 6.563}, {'end': 4678.6, 'text': 'And this is actually giving a 93% accuracy on that task in about 15 minutes.', 'start': 4672.337, 'duration': 6.263}, {'end': 4688.083, 'text': 'on the IMDB dataset which contains thousands of full-length movie reviews, like 1,000 to 3,000 word movie reviews.', 'start': 4680.718, 'duration': 7.365}, {'end': 4696.388, 'text': 'And this number here that we got with the same three lines of code would have been the best in the world for this task in a very, very,', 'start': 4688.743, 'duration': 7.645}, {'end': 4700.851, 'text': 'very popular academics dataset in like 2015, I think.', 'start': 4696.388, 'duration': 4.463}, {'end': 4707.835, 'text': 'So we are creating world-class models in our browser using the same basic code.', 'start': 4701.651, 'duration': 6.184}, {'end': 4719.087, 'text': "Here's the same basic steps again from import star untar data tabular data loaders from csv learner fit.", 'start': 4711.544, 'duration': 7.543}], 'summary': 'Using the same basic code, achieving 93% accuracy in 15 minutes on sentiment analysis of imdb dataset, which was world-class in 2015.', 'duration': 82.746, 'max_score': 4636.341, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/_QUEXsHfsA0/pics/_QUEXsHfsA04636341.jpg'}, {'end': 4821.345, 'src': 'embed', 'start': 4794.061, 'weight': 5, 'content': [{'end': 4806.352, 'text': 'if only you have a way to parameterize a model and you have an update procedure which can update the weights to make you better at your loss function,', 'start': 4794.061, 'duration': 12.291}, {'end': 4814.719, 'text': 'and in this case we can use neural networks, which are totally flexible functions.', 'start': 4806.352, 'duration': 8.367}, {'end': 4818.104, 'text': "so um, That's it for this first lesson.", 'start': 4814.719, 'duration': 3.385}, {'end': 4821.345, 'text': "It's a little bit shorter than other lessons are going to be.", 'start': 4818.164, 'duration': 3.181}], 'summary': 'Parameterize models, update weights with neural networks for flexibility. short first lesson.', 'duration': 27.284, 'max_score': 4794.061, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/_QUEXsHfsA0/pics/_QUEXsHfsA04794061.jpg'}, {'end': 4894.078, 'src': 'embed', 'start': 4846.686, 'weight': 6, 'content': [{'end': 4856.491, 'text': 'So what I suggest you do over the next week before you work on the next lesson, is just make sure that you can spin up a GPU server,', 'start': 4846.686, 'duration': 9.805}, {'end': 4862.092, 'text': "that you can shut it down when it's finished, that you can run all of the the code here.", 'start': 4856.491, 'duration': 5.601}, {'end': 4867.077, 'text': 'And as you go through it, see, you know, is this using Python in a way you recognize?', 'start': 4862.593, 'duration': 4.484}, {'end': 4869.54, 'text': 'Use the documentation.', 'start': 4867.097, 'duration': 2.443}, {'end': 4870.941, 'text': 'use that doc function.', 'start': 4869.54, 'duration': 1.401}, {'end': 4876.447, 'text': 'do some searching of the fastai doc, see what it does.', 'start': 4870.941, 'duration': 5.506}, {'end': 4882.131, 'text': 'See if you can actually grab the fastai Documentation notebooks themselves and run them.', 'start': 4876.467, 'duration': 5.664}, {'end': 4889.455, 'text': 'just try to get comfortable That you kind of can know your way around, Because the most important thing to do with this style of learning,', 'start': 4882.131, 'duration': 7.324}, {'end': 4894.078, 'text': 'this top-down learning, is to be able to run experiments, And that means you need to be able to run code.', 'start': 4889.455, 'duration': 4.623}], 'summary': 'Practice running code using gpu server, fastai documentation, and recognize python usage.', 'duration': 47.392, 'max_score': 4846.686, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/_QUEXsHfsA0/pics/_QUEXsHfsA04846686.jpg'}, {'end': 4950.407, 'src': 'embed', 'start': 4931.079, 'weight': 8, 'content': [{'end': 4948.486, 'text': 'where we will learn about transfer learning and then we will move on to creating a an actual production version of an application that we can actually put out on the internet and you can start building apps that you can show your friends and they can start playing with.', 'start': 4931.079, 'duration': 17.407}, {'end': 4950.407, 'text': 'All right, bye everybody.', 'start': 4949.286, 'duration': 1.121}], 'summary': 'Learn transfer learning and create a production app for internet deployment.', 'duration': 19.328, 'max_score': 4931.079, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/_QUEXsHfsA0/pics/_QUEXsHfsA04931079.jpg'}], 'start': 4445.507, 'title': 'Python programming, fast.ai, and neural networks', 'summary': 'Covers customized python programming for data science, highlighting 40 years of programming experience, positive student feedback, fast.ai library for creating world-class models, and introduction to neural networks in the context of a global pandemic.', 'chapters': [{'end': 4555.233, 'start': 4445.507, 'title': 'Customized python programming for data science', 'summary': "Highlights the speaker's 40 years of programming experience, the unique coding practices used in the course, and the positive feedback from students, resulting in improved understanding of python coding and software engineering.", 'duration': 109.726, 'highlights': ['The speaker has 40 years of programming experience, emphasizing the use of unique coding practices and the ability to understand code even after years.', 'Many students from previous courses have learned a lot about coding, Python coding, and software engineering from the course.', "The speaker's coding style incorporates ideas from various languages and notations, heavily customized for data science, which may not align with traditional Pythonic approaches.", 'The speaker recommends fitting in with organizational programming practices rather than following the customized approach used in the course.', 'The speaker mentions using programming practices that may not align with traditional Python programming approaches due to experience with various languages.']}, {'end': 4768.197, 'start': 4556.314, 'title': 'Fast.ai: creating world-class models', 'summary': "Showcases the fast.ai library's ability to rapidly create world-class models for various tasks, such as segmentation, sentiment analysis, tabular data prediction, and collaborative filtering, achieving high accuracy in a short time using the same basic code.", 'duration': 211.883, 'highlights': ['The Fast.ai library allows for rapid creation of models for various tasks, such as segmentation, sentiment analysis, tabular data prediction, and collaborative filtering, achieving high accuracy in a short time using the same basic code.', 'The model for segmentation successfully identifies objects in images, achieving its results in under 20 seconds and reaching near perfection with only a few minutes of training.', 'The sentiment analysis model achieves 93% accuracy on the IMDB dataset, which was considered world-class in 2015, using the same three lines of code.', 'The Fast.ai library enables the creation of models for tabular data prediction and collaborative filtering, showcasing its versatility in handling different types of data and tasks.']}, {'end': 4950.407, 'start': 4769.417, 'title': 'Introduction to neural networks', 'summary': 'Introduces the basic concepts of neural networks and emphasizes the importance of being able to run code and experiment with the provided materials before moving on to the next lesson, due to the ongoing global pandemic.', 'duration': 180.99, 'highlights': ["Arthur Samuel's basic description of parameterizing a model and using an update procedure allows for the flexibility of neural networks, enabling various applications.", 'The importance of being able to run code and experiment is emphasized, as it is crucial for top-down learning and practical application of the concepts.', 'Recommendation to ensure the ability to spin up a GPU server, run the provided code, get familiar with Python, and utilize the fastai documentation and notebooks for learning.', 'Emphasis on the need to read the chapter of the book, go through the questionnaire, and prepare for learning about validation sets, test sets, and transfer learning in the upcoming lessons.', 'Upcoming topics include transfer learning and creating a production version of an application for deployment on the internet, enabling the audience to build and showcase their own apps.']}], 'duration': 504.9, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/_QUEXsHfsA0/pics/_QUEXsHfsA04445507.jpg', 'highlights': ['The Fast.ai library enables the creation of models for various tasks, achieving high accuracy in a short time using the same basic code.', 'The sentiment analysis model achieves 93% accuracy on the IMDB dataset, considered world-class in 2015, using the same three lines of code.', 'The model for segmentation successfully identifies objects in images, achieving near perfection with only a few minutes of training.', 'The speaker has 40 years of programming experience, emphasizing the use of unique coding practices and the ability to understand code even after years.', 'Many students from previous courses have learned a lot about coding, Python coding, and software engineering from the course.', "Arthur Samuel's basic description of parameterizing a model and using an update procedure allows for the flexibility of neural networks, enabling various applications.", 'The importance of being able to run code and experiment is emphasized, crucial for top-down learning and practical application of the concepts.', 'Recommendation to ensure the ability to spin up a GPU server, run the provided code, get familiar with Python, and utilize the fastai documentation and notebooks for learning.', 'Upcoming topics include transfer learning and creating a production version of an application for deployment on the internet, enabling the audience to build and showcase their own apps.']}], 'highlights': ['The live launch of Deep Learning for Coders Lesson 1 during a pandemic', 'Development of Fast AI Library Version 2 and a peer-reviewed paper', 'The syllabus is closely based on a book and is available for free as Jupyter notebooks', 'Different versions of the notebooks are available, including a full notebook with prose, pictures, and everything', 'Sylvain and Rachel are co-authors of the Fast.ai library and are experts in math and data ethics respectively, providing valuable insights throughout the course.', 'Deep learning is a useful tool in many areas and is equivalent to, or better than, human performance in certain tasks, showcasing its wide applicability and effectiveness. Demonstrated superiority over human performance in various tasks.', 'The breakthrough in the 1980s showed that adding a second layer of neurons made it mathematically provable that any mathematical model could be approximated to any level of accuracy with neural networks, signifying a significant advancement in the capabilities of neural networks. Mathematically proven capability to approximate any mathematical model to any level of accuracy with neural networks.', 'The potential of neural nets enables capabilities such as perceiving, recognizing, and identifying surroundings without human training or control.', 'The fast AI approach emphasizes training state-of-the-art world class models, ensuring a practical and comprehensive learning experience.', 'PyTorch usage has increased from 20% to 80% in the last 12 months, while TensorFlow usage has decreased from 80% to 20%, indicating its growing dominance in the field.', "Jupyter Notebook's REPL system offers features such as headings, graphical outputs, and multimedia, making it a widely used and powerful system for teaching and development.", 'The chapter introduces the concept of Jupyter Notebook, explaining its REPL-based system, two modes (edit and command), keyboard shortcuts, and its usefulness in creating formatted text, plots, and lists.', 'Trained model achieves 99.96% accuracy in recognizing cats from dogs.', 'Introduction of Stochastic Gradient Descent as a method to update weights based on performance.', 'The function that sits in the middle of the architecture is what we call an architecture, which is adjusted by the parameters to produce predictions.', 'The predictions, based on independent and dependent variables, are measured by the loss function, which is used to update the parameters.', 'Lack of labeled data is a common challenge in deep learning, as organizations often struggle to obtain enough labeled data for training models.', 'Black people in the US are arrested seven times more often than whites for marijuana possession, despite similar marijuana use, leading to biased data and predictions.', 'The Fast.ai library enables the creation of models for various tasks, achieving high accuracy in a short time using the same basic code.', 'The sentiment analysis model achieves 93% accuracy on the IMDB dataset, considered world-class in 2015, using the same three lines of code.', 'The model for segmentation successfully identifies objects in images, achieving near perfection with only a few minutes of training.', 'The speaker has 40 years of programming experience, emphasizing the use of unique coding practices and the ability to understand code even after years.', 'Many students from previous courses have learned a lot about coding, Python coding, and software engineering from the course.', "Arthur Samuel's basic description of parameterizing a model and using an update procedure allows for the flexibility of neural networks, enabling various applications.", 'Recommendation to ensure the ability to spin up a GPU server, run the provided code, get familiar with Python, and utilize the fastai documentation and notebooks for learning.']}