title
Lesson 1: Deep Learning 2018

description
NB: Please go to http://course.fast.ai to view this video since there is important updated information there. If you have questions, use the forums at http://forums.fast.ai Welcome to the start of your fast.ai journey! In today’s lesson you’ll set up your deep learning server, and training your first image classification model (a convolutional neural network, or CNN), which will learn to distinguish dogs from cats nearly perfectly. If you need help at any time, head over to forums.fast.ai where over a thousand students are discussing the course and have provided lots of tips and tricks for you.

detail
{'title': 'Lesson 1: Deep Learning 2018', 'heatmap': [{'end': 792.212, 'start': 517.668, 'weight': 0.791}, {'end': 1200.563, 'start': 1081.819, 'weight': 0.703}, {'end': 3748.097, 'start': 3681.83, 'weight': 0.726}], 'summary': "Course 'lesson 1: deep learning 2018' covers practical deep learning for coders, gpu setup, ubuntu 1604 and fastai environment setup, image classification and model training achieving 99% accuracy in 17 seconds, visualizing image classifier model, top-down learning approach, neural network for philosophy and image recognition, gpu's impact on deep learning adoption by major companies, concepts of convolutional neural network (cnn) and its application in edge detection and image transformation, and deep learning network evolution and model training techniques.", 'chapters': [{'end': 490.881, 'segs': [{'end': 27.303, 'src': 'embed', 'start': 0.899, 'weight': 0, 'content': [{'end': 4.862, 'text': 'Hi everybody welcome to practical deep learning for coders.', 'start': 0.899, 'duration': 3.963}, {'end': 9.286, 'text': 'This is part one of our two-part course.', 'start': 5.163, 'duration': 4.123}, {'end': 16.472, 'text': "I'm Presenting this from the Data Institute in San Francisco.", 'start': 9.286, 'duration': 7.186}, {'end': 20.356, 'text': "We'll be doing seven lessons in this part of the course.", 'start': 16.472, 'duration': 3.884}, {'end': 22.417, 'text': 'Most of them will be about a couple of hours long.', 'start': 20.356, 'duration': 2.061}, {'end': 27.303, 'text': 'this first one may be a little bit shorter and Practical.', 'start': 22.417, 'duration': 4.886}], 'summary': 'Part one: practical deep learning course in san francisco with 7 lessons, most 2 hours long.', 'duration': 26.404, 'max_score': 0.899, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/IPBSB1HLNLo/pics/IPBSB1HLNLo899.jpg'}, {'end': 73.039, 'src': 'embed', 'start': 48.612, 'weight': 1, 'content': [{'end': 55.967, 'text': "We're learning about from scratch, and Now I should mention that our videos are hosted on YouTube,", 'start': 48.612, 'duration': 7.355}, {'end': 61.17, 'text': 'but we strongly recommend watching them via our website at course.fast.ai.', 'start': 55.967, 'duration': 5.203}, {'end': 64.132, 'text': "Although they're exactly the same videos.", 'start': 62.211, 'duration': 1.921}, {'end': 73.039, 'text': "the important thing about watching them through our website is that you'll get all of the information you need about updates to libraries,", 'start': 64.132, 'duration': 8.907}], 'summary': 'Learn from scratch at course.fast.ai for the latest updates.', 'duration': 24.427, 'max_score': 48.612, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/IPBSB1HLNLo/pics/IPBSB1HLNLo48612.jpg'}, {'end': 152.092, 'src': 'embed', 'start': 118.975, 'weight': 2, 'content': [{'end': 120.676, 'text': 'On forums.fast.ai.', 'start': 118.975, 'duration': 1.701}, {'end': 126.718, 'text': 'there are thousands of other learners talking about every lesson and lots of other topics besides.', 'start': 120.676, 'duration': 6.042}, {'end': 130.139, 'text': "It's the most active deep learning community on the internet by far.", 'start': 126.718, 'duration': 3.421}, {'end': 134.621, 'text': 'so definitely, Register there and start getting involved.', 'start': 130.139, 'duration': 4.482}, {'end': 140.123, 'text': "you'll get a lot more out of this course if you do that.", 'start': 134.621, 'duration': 5.502}, {'end': 143.284, 'text': "So we're going to start by doing some coding,", 'start': 140.123, 'duration': 3.161}, {'end': 152.092, 'text': "and This is an approach we're going to be talking about in a moment called the top-down approach to study, but let's learn it by doing it.", 'start': 143.284, 'duration': 8.808}], 'summary': 'Join forums.fast.ai for active community discussions and top-down learning approach.', 'duration': 33.117, 'max_score': 118.975, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/IPBSB1HLNLo/pics/IPBSB1HLNLo118975.jpg'}, {'end': 222.356, 'src': 'embed', 'start': 197.045, 'weight': 3, 'content': [{'end': 207.252, 'text': 'CUDA is the language and framework that nearly all deep learning libraries and practitioners use to do their work.', 'start': 197.045, 'duration': 10.207}, {'end': 213.556, 'text': "Obviously, it's not ideal that we're stuck with one particular vendors cards, and over time We hope to see more competition in this space.", 'start': 207.252, 'duration': 6.304}, {'end': 222.356, 'text': "but for now we do need an Nvidia GPU, and Your laptop almost certainly doesn't have one, unless you specifically went out of your way to buy,", 'start': 213.556, 'duration': 8.8}], 'summary': 'Cuda is widely used in deep learning, but more competition is desired. nvidia gpus are currently necessary.', 'duration': 25.311, 'max_score': 197.045, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/IPBSB1HLNLo/pics/IPBSB1HLNLo197045.jpg'}, {'end': 320.966, 'src': 'embed', 'start': 281.264, 'weight': 4, 'content': [{'end': 286.29, 'text': 'Jupiter notebook, in a recent survey of tens of thousands of data scientists,', 'start': 281.264, 'duration': 5.026}, {'end': 291.595, 'text': 'was rated as the third most important tool in the data scientist toolbox.', 'start': 286.29, 'duration': 5.305}, {'end': 296.66, 'text': "It's really important that you get to learn it well and all of our courses will be run through Jupiter.", 'start': 291.595, 'duration': 5.065}, {'end': 297.261, 'text': 'Yes, Rachel.', 'start': 296.66, 'duration': 0.601}, {'end': 298.322, 'text': 'You have a question or a comment?', 'start': 297.281, 'duration': 1.041}, {'end': 307.579, 'text': "Oh, I just wanted to point out that you get, I believe, 10 free hours, so if you wanted to try Crestle out, you're not having to pay right away.", 'start': 298.635, 'duration': 8.944}, {'end': 314.643, 'text': 'Yeah He might have changed that recently to less hours, but you can check the FAQ or the pricing, but you certainly get some free hours.', 'start': 307.659, 'duration': 6.984}, {'end': 320.966, 'text': "The pricing varies because this actually runs on top of Amazon Web Services, so at the moment it's $0.60 an hour.", 'start': 315.603, 'duration': 5.363}], 'summary': 'Jupyter notebook is the third most important tool for data scientists, offering around 10 free hours for usage.', 'duration': 39.702, 'max_score': 281.264, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/IPBSB1HLNLo/pics/IPBSB1HLNLo281264.jpg'}, {'end': 445.836, 'src': 'embed', 'start': 422.346, 'weight': 5, 'content': [{'end': 431.357, 'text': "for 40 cents an hour, so it's cheaper than Cressel I get a machine that's actually going to be much faster than Cressel 60 cent an hour machine.", 'start': 422.346, 'duration': 9.011}, {'end': 435.114, 'text': 'or for 65 cents an hour, Way, way, way faster, right?', 'start': 431.357, 'duration': 3.757}, {'end': 440.775, 'text': "So I'm going to actually show you how to get started with with the with the paper space approach,", 'start': 435.434, 'duration': 5.341}, {'end': 445.836, 'text': 'Because that actually is going to do everything from scratch.', 'start': 440.775, 'duration': 5.061}], 'summary': "Paper space offers faster machines for 40-65 cents an hour, making it cheaper and more efficient than cressel's 60-cent-an-hour machine.", 'duration': 23.49, 'max_score': 422.346, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/IPBSB1HLNLo/pics/IPBSB1HLNLo422346.jpg'}], 'start': 0.899, 'title': 'Practical deep learning for coders and setting up gpu for neural network training', 'summary': 'Introduces a two-part course on practical deep learning for coders, covering seven lessons, emphasizing a coding-focused approach to learning deep learning. it also discusses the importance of using a gpu for training neural networks and recommends options like crestle and paper space, highlighting their pricing and performance advantages over cpu.', 'chapters': [{'end': 143.284, 'start': 0.899, 'title': 'Practical deep learning for coders', 'summary': 'Introduces a two-part course on practical deep learning for coders, covering seven lessons, emphasizing a coding-focused approach to learning deep learning, and recommending viewers to access the videos and resources through the course website at course.fast.ai.', 'duration': 142.385, 'highlights': ['The course covers seven lessons and emphasizes a coding-focused approach to learning deep learning. The course consists of seven lessons, most of which will be about a couple of hours long, and is focused on a coding-centric approach to learning deep learning.', 'The presenter recommends watching the videos through the course website at course.fast.ai to access necessary information about updates and resources. The presenter strongly recommends watching the videos via the website at course.fast.ai to access information about updates to libraries, biolocations, FAQs, and more.', 'The presenter advises viewers to engage with the active deep learning community at forums.fast.ai for assistance and involvement. The presenter recommends registering on forums.fast.ai, as it hosts an active deep learning community where learners can seek help, discuss lessons, and engage in various topics related to deep learning.']}, {'end': 490.881, 'start': 143.284, 'title': 'Setting up gpu for neural network training', 'summary': 'Discusses the importance of using a gpu for training neural networks, recommending options like crestle and paper space, and highlighting their pricing and performance advantages over cpu. it also emphasizes the significance of jupyter notebook in data science and its integration with the course.', 'duration': 347.597, 'highlights': ['Using a GPU for training neural networks is essential, with Nvidia GPUs being the preferred choice due to their support for CUDA, the primary language and framework for deep learning, and the availability of affordable rental options like Crestle and Paper Space. (Relevance: 5)', 'Crestle offers easy access to a GPU-based computer, providing a simple setup through a Jupiter notebook and 10 free hours for trial, with pricing at $0.60 per hour and the flexibility to enable/disable GPU to manage costs. (Relevance: 4)', 'Paper Space presents a cost-effective alternative to Crestle, offering faster machines at 40-65 cents per hour, with the option to choose data centers and storage, emphasizing the importance of minimizing machine starts to avoid unnecessary storage costs. (Relevance: 3)', "Emphasizes the significance of Jupyter notebook in data science, highlighting its ranking as the third most important tool in the data scientist toolbox and its integral role in the course's curriculum. (Relevance: 2)", 'Recommends checking the FAQ or pricing for updated information on free hours and costs for using Crestle, with the mention of potential changes in the free trial duration and the clarification that Crestle runs on top of Amazon Web Services. (Relevance: 1)']}], 'duration': 489.982, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/IPBSB1HLNLo/pics/IPBSB1HLNLo899.jpg', 'highlights': ['The course consists of seven lessons, most of which will be about a couple of hours long, and is focused on a coding-centric approach to learning deep learning.', 'The presenter strongly recommends watching the videos via the website at course.fast.ai to access information about updates to libraries, biolocations, FAQs, and more.', 'The presenter recommends registering on forums.fast.ai, as it hosts an active deep learning community where learners can seek help, discuss lessons, and engage in various topics related to deep learning.', 'Using a GPU for training neural networks is essential, with Nvidia GPUs being the preferred choice due to their support for CUDA, the primary language and framework for deep learning, and the availability of affordable rental options like Crestle and Paper Space.', 'Crestle offers easy access to a GPU-based computer, providing a simple setup through a Jupiter notebook and 10 free hours for trial, with pricing at $0.60 per hour and the flexibility to enable/disable GPU to manage costs.', 'Paper Space presents a cost-effective alternative to Crestle, offering faster machines at 40-65 cents per hour, with the option to choose data centers and storage, emphasizing the importance of minimizing machine starts to avoid unnecessary storage costs.', "Emphasizes the significance of Jupyter notebook in data science, highlighting its ranking as the third most important tool in the data scientist toolbox and its integral role in the course's curriculum.", 'Recommends checking the FAQ or pricing for updated information on free hours and costs for using Crestle, with the mention of potential changes in the free trial duration and the clarification that Crestle runs on top of Amazon Web Services.']}, {'end': 926.191, 'segs': [{'end': 792.212, 'src': 'heatmap', 'start': 490.881, 'weight': 0, 'content': [{'end': 506.624, 'text': 'and The only other thing you need to do is turn on public IP so that we can actually log into this and We can turn off auto snapshot to save the money of not having backups.', 'start': 490.881, 'duration': 15.743}, {'end': 508.605, 'text': 'All right.', 'start': 506.624, 'duration': 1.981}, {'end': 517.668, 'text': 'so if you then click on create your paper space about a minute later, you will find That your machine will pop up.', 'start': 508.605, 'duration': 9.063}, {'end': 528.324, 'text': 'here is my Ubuntu 1604 machine, and If you check your email, you will find that they have emailed you a password.', 'start': 517.668, 'duration': 10.656}, {'end': 534.827, 'text': 'So you can copy that, and you can go to your machine and enter your password.', 'start': 529.004, 'duration': 5.823}, {'end': 542.011, 'text': 'Now to paste the password, you would press Ctrl-Shift-V, or on Mac I guess Apple-Shift-V.', 'start': 535.368, 'duration': 6.643}, {'end': 547.454, 'text': "So it's slightly different to normal pasting, or of course you can just type it in.", 'start': 543.572, 'duration': 3.882}, {'end': 551.496, 'text': 'And here we are.', 'start': 551.016, 'duration': 0.48}, {'end': 554.78, 'text': 'Now we can make a little bit more room here by clicking on these little arrows.', 'start': 552.059, 'duration': 2.721}, {'end': 557.201, 'text': 'I can zoom in a little bit.', 'start': 556.201, 'duration': 1}, {'end': 565.985, 'text': "And so as you can see, we've got like a terminal that's sitting inside our browser, which is kind of quite a handy way to do it.", 'start': 558.221, 'duration': 7.764}, {'end': 569.446, 'text': 'So now we need to configure this for the course.', 'start': 566.445, 'duration': 3.001}, {'end': 589.981, 'text': 'And so the way you configure it for the course is you type curl http://files Fast AI slash setup, slash paperspace, Pipe bash.', 'start': 569.986, 'duration': 19.995}, {'end': 602.816, 'text': "Okay. and so that's then going to run a script which is going to set up all of the CUDA drivers and the special Python Reaper Python distribution we use,", 'start': 589.981, 'duration': 12.835}, {'end': 610.04, 'text': 'called Anaconda, all of the libraries, all of the courses And the data we use for the first part of the course.', 'start': 602.816, 'duration': 7.224}, {'end': 618.545, 'text': "Okay, so that takes an hour or so and when it's finished running, you'll need to reboot your computer.", 'start': 610.04, 'duration': 8.505}, {'end': 621.806, 'text': 'So to reboot not your own computer but your papers-based computer.', 'start': 618.545, 'duration': 3.261}, {'end': 626.129, 'text': 'and so to do that, You can just click on this little circular restart machine button.', 'start': 621.806, 'duration': 4.323}, {'end': 629.954, 'text': "Okay, so When it comes back up, you'll be ready to go.", 'start': 626.149, 'duration': 3.805}, {'end': 635.755, 'text': "so what you'll find Is that you've now got an anaconda3 directory.", 'start': 629.954, 'duration': 5.801}, {'end': 637.795, 'text': "that's where your Python is.", 'start': 635.755, 'duration': 2.04}, {'end': 644.697, 'text': "you've got a data directory Which contains the data for the first part of this course first lesson, which is that dogs and cats?", 'start': 637.795, 'duration': 6.902}, {'end': 653.318, 'text': "and you've got a fast AI directory, and That contains everything for this course.", 'start': 645.757, 'duration': 7.561}, {'end': 667.104, 'text': 'So what you should do is CD fast AI and from time to time you should go get pull and that will just make sure that all of your Fast AI stuff is up to date.', 'start': 653.318, 'duration': 13.786}, {'end': 671.325, 'text': 'And also, from time to time you might want to just check that your Python libraries are up to date,', 'start': 667.104, 'duration': 4.221}, {'end': 676.947, 'text': 'And so you can type Conda and update to do that All right.', 'start': 671.325, 'duration': 5.622}, {'end': 683.168, 'text': "so make sure that you've CDD into fast AI and then you can type Jupiter notebook.", 'start': 676.947, 'duration': 6.221}, {'end': 688.43, 'text': 'Right there it is.', 'start': 687.469, 'duration': 0.961}, {'end': 694.734, 'text': 'so we now have a jupyter notebook serving it running and we want to connect that, And so you can see here.', 'start': 688.43, 'duration': 6.304}, {'end': 699.637, 'text': 'It says copy, paste this URL Into your browser when you connect.', 'start': 694.754, 'duration': 4.883}, {'end': 706.582, 'text': 'so if you double click on it, then that will actually.', 'start': 699.637, 'duration': 6.945}, {'end': 711.122, 'text': 'That will actually copy it for you and You can go and paste it.', 'start': 706.582, 'duration': 4.54}, {'end': 716.983, 'text': 'but you need to change this localhost To be the paper space IP address.', 'start': 711.122, 'duration': 5.861}, {'end': 721.364, 'text': 'so if you click on the little arrows to go smaller, You can see the IP address is here.', 'start': 716.983, 'duration': 4.381}, {'end': 728.446, 'text': "so I'll just copy that and paste it Where it used to say localhost.", 'start': 721.364, 'duration': 7.082}, {'end': 734.848, 'text': "okay, so it's now HTTP and then my IP and then everything else I copied before.", 'start': 728.446, 'duration': 6.402}, {'end': 735.428, 'text': 'and So there it is.', 'start': 734.848, 'duration': 0.58}, {'end': 748.4, 'text': 'So this is the FastAI Git repo and our courses are all in Courses and in there the Deep Learning Part 1 is DL1,', 'start': 736.909, 'duration': 11.491}, {'end': 760.193, 'text': 'and in there you will find Lesson1.ipynb.', 'start': 748.4, 'duration': 11.793}, {'end': 764.816, 'text': "Here we are ready to go, depending whether you're using Crestle or paper space or something else.", 'start': 760.193, 'duration': 4.623}, {'end': 766.057, 'text': 'if you check courses, stop fast.', 'start': 764.816, 'duration': 1.241}, {'end': 771.76, 'text': 'today I will keep putting additional videos and links to information about how to set up other.', 'start': 766.057, 'duration': 5.703}, {'end': 777.864, 'text': 'You know good, you put a notebook providers as well.', 'start': 771.76, 'duration': 6.104}, {'end': 789.131, 'text': "So to run a cell in Jupiter notebook, you select the cell and you hold down, shift and press enter, and If you've got the toolbar showing,", 'start': 777.864, 'duration': 11.267}, {'end': 792.212, 'text': 'you can just click on the little run button.', 'start': 789.131, 'duration': 3.081}], 'summary': 'Set up and configure a paperspace, run fastai setup, and connect to jupyter notebook for deep learning part 1 course.', 'duration': 60.615, 'max_score': 490.881, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/IPBSB1HLNLo/pics/IPBSB1HLNLo490881.jpg'}, {'end': 610.04, 'src': 'embed', 'start': 569.986, 'weight': 4, 'content': [{'end': 589.981, 'text': 'And so the way you configure it for the course is you type curl http://files Fast AI slash setup, slash paperspace, Pipe bash.', 'start': 569.986, 'duration': 19.995}, {'end': 602.816, 'text': "Okay. and so that's then going to run a script which is going to set up all of the CUDA drivers and the special Python Reaper Python distribution we use,", 'start': 589.981, 'duration': 12.835}, {'end': 610.04, 'text': 'called Anaconda, all of the libraries, all of the courses And the data we use for the first part of the course.', 'start': 602.816, 'duration': 7.224}], 'summary': "Running the script 'curl http://files fast ai slash setup, slash paperspace, pipe bash' sets up cuda drivers, anaconda, libraries, and data for the course.", 'duration': 40.054, 'max_score': 569.986, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/IPBSB1HLNLo/pics/IPBSB1HLNLo569986.jpg'}, {'end': 667.104, 'src': 'embed', 'start': 635.755, 'weight': 5, 'content': [{'end': 637.795, 'text': "that's where your Python is.", 'start': 635.755, 'duration': 2.04}, {'end': 644.697, 'text': "you've got a data directory Which contains the data for the first part of this course first lesson, which is that dogs and cats?", 'start': 637.795, 'duration': 6.902}, {'end': 653.318, 'text': "and you've got a fast AI directory, and That contains everything for this course.", 'start': 645.757, 'duration': 7.561}, {'end': 667.104, 'text': 'So what you should do is CD fast AI and from time to time you should go get pull and that will just make sure that all of your Fast AI stuff is up to date.', 'start': 653.318, 'duration': 13.786}], 'summary': 'The python data directory contains the course material for dogs and cats. cd into fast ai and periodically update with git pull to keep the content up to date.', 'duration': 31.349, 'max_score': 635.755, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/IPBSB1HLNLo/pics/IPBSB1HLNLo635755.jpg'}, {'end': 805.419, 'src': 'embed', 'start': 777.864, 'weight': 6, 'content': [{'end': 789.131, 'text': "So to run a cell in Jupiter notebook, you select the cell and you hold down, shift and press enter, and If you've got the toolbar showing,", 'start': 777.864, 'duration': 11.267}, {'end': 792.212, 'text': 'you can just click on the little run button.', 'start': 789.131, 'duration': 3.081}, {'end': 800.714, 'text': "so you'll notice that some cells contain Code and some contain text, and some contain pictures and some contain videos.", 'start': 792.212, 'duration': 8.502}, {'end': 805.419, 'text': "so this environment basically has know it's.", 'start': 800.714, 'duration': 4.705}], 'summary': 'In jupyter notebook, to run a cell, use shift+enter or the run button. cells can contain code, text, pictures, or videos.', 'duration': 27.555, 'max_score': 777.864, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/IPBSB1HLNLo/pics/IPBSB1HLNLo777864.jpg'}, {'end': 888.816, 'src': 'embed', 'start': 860.084, 'weight': 7, 'content': [{'end': 864.487, 'text': "This is just a side note, but I wanted to point out that we're using Python 3 here.", 'start': 860.084, 'duration': 4.403}, {'end': 865.508, 'text': 'Yes, thank you.', 'start': 864.628, 'duration': 0.88}, {'end': 865.969, 'text': 'Python 3.6.', 'start': 865.608, 'duration': 0.361}, {'end': 869.432, 'text': "And so you may get some errors if you're still using Python 2.", 'start': 865.969, 'duration': 3.463}, {'end': 873.695, 'text': 'Yeah And it is important to switch to Python 3.', 'start': 869.432, 'duration': 4.263}, {'end': 878.439, 'text': 'You know, now, well, for Fast.ai, you require it.', 'start': 873.695, 'duration': 4.744}, {'end': 885.734, 'text': 'But, you know, increasingly, a lot of libraries are removing support for Python 2.', 'start': 879.08, 'duration': 6.654}, {'end': 888.816, 'text': 'Thanks, Rachel.', 'start': 885.734, 'duration': 3.082}], 'summary': 'Switch to python 3.6 for fast.ai and other libraries.', 'duration': 28.732, 'max_score': 860.084, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/IPBSB1HLNLo/pics/IPBSB1HLNLo860084.jpg'}], 'start': 490.881, 'title': 'Setting up ubuntu 1604 and fastai environment', 'summary': 'Provides a step-by-step guide on setting up an ubuntu 1604 machine on paperspace, including turning on public ip, disabling auto snapshot, and receiving the password via email. it also demonstrates setting up the fastai environment using a web-based terminal, configuring it for the course, and accessing course materials and data, involving running scripts, updating libraries, and connecting to jupyter notebook. the process takes approximately a minute for setting up ubuntu 1604, and entails configuring for the course and accessing course materials for fastai environment.', 'chapters': [{'end': 551.496, 'start': 490.881, 'title': 'Setting up ubuntu 1604 on paperspace', 'summary': 'Provides a step-by-step guide on setting up an ubuntu 1604 machine on paperspace, including turning on public ip, disabling auto snapshot, and receiving the password via email, with the unique method of pasting the password highlighted. the process takes approximately a minute.', 'duration': 60.615, 'highlights': ['Upon creating a PaperSpace machine, an Ubuntu 1604 machine will appear in about a minute.', 'Turning on public IP is essential for logging into the machine.', 'Disabling auto snapshot is recommended to save on backup costs.', 'The password for the machine is emailed, and the unique method of pasting the password using Ctrl-Shift-V on Windows or Apple-Shift-V on Mac is explained.']}, {'end': 926.191, 'start': 552.059, 'title': 'Setting up fastai environment', 'summary': 'Demonstrates setting up the fastai environment using a web-based terminal, configuring it for the course, and accessing course materials and data, involving running scripts, updating libraries, and connecting to jupyter notebook, emphasizing the importance of using python 3.', 'duration': 374.132, 'highlights': ["Setting up the FastAI environment involves running a script using 'curl' to configure CUDA drivers, Anaconda Python distribution, libraries, and course data, which takes about an hour to complete. The process of setting up the FastAI environment includes running a script using 'curl' to configure CUDA drivers, Anaconda Python distribution, libraries, and course data, which takes about an hour to complete.", "After setting up, accessing the course materials involves accessing 'anaconda3' directory for Python, 'data' directory for course data, and 'fast AI' directory for course content and updating the Fast AI content using 'git pull'. Accessing the course materials involves accessing 'anaconda3' directory for Python, 'data' directory for course data, and 'fast AI' directory for course content and updating the Fast AI content using 'git pull'.", "Connecting to Jupyter notebook requires adjusting the URL by replacing 'localhost' with the paper space IP address and running cells in Jupyter notebook using 'shift+enter' or the toolbar, with emphasis on using Python 3 for Fast.ai. Connecting to Jupyter notebook requires adjusting the URL by replacing 'localhost' with the paper space IP address and running cells in Jupyter notebook using 'shift+enter' or the toolbar, with emphasis on using Python 3 for Fast.ai.", 'Emphasizing the importance of using Python 3 for Fast.ai and the increasing trend of libraries removing support for Python 2. Emphasizing the importance of using Python 3 for Fast.ai and the increasing trend of libraries removing support for Python 2.']}], 'duration': 435.31, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/IPBSB1HLNLo/pics/IPBSB1HLNLo490881.jpg', 'highlights': ['Setting up an Ubuntu 1604 machine on PaperSpace takes approximately a minute.', 'Turning on public IP is essential for logging into the machine.', 'Disabling auto snapshot is recommended to save on backup costs.', 'The password for the machine is emailed, and the unique method of pasting the password using Ctrl-Shift-V on Windows or Apple-Shift-V on Mac is explained.', "Setting up the FastAI environment involves running a script using 'curl' to configure CUDA drivers, Anaconda Python distribution, libraries, and course data, which takes about an hour to complete.", "Accessing the course materials involves accessing 'anaconda3' directory for Python, 'data' directory for course data, and 'fast AI' directory for course content and updating the Fast AI content using 'git pull'.", "Connecting to Jupyter notebook requires adjusting the URL by replacing 'localhost' with the PaperSpace IP address and running cells in Jupyter notebook using 'shift+enter' or the toolbar, with emphasis on using Python 3 for Fast.ai.", 'Emphasizing the importance of using Python 3 for Fast.ai and the increasing trend of libraries removing support for Python 2.']}, {'end': 1576.894, 'segs': [{'end': 1007.199, 'src': 'embed', 'start': 977.569, 'weight': 4, 'content': [{'end': 981.95, 'text': "if you're not familiar with the idea of training sets and validation sets,", 'start': 977.569, 'duration': 4.381}, {'end': 987.512, 'text': 'It would be a very good idea to check out our practical machine learning course,', 'start': 981.95, 'duration': 5.562}, {'end': 995.435, 'text': 'Which tells you a lot about this kind of stuff of like that the basics of how to set up and run? machine learning projects more generally.', 'start': 987.512, 'duration': 7.923}, {'end': 1000.45, 'text': 'Would you recommend that people take that course before this one?', 'start': 997.025, 'duration': 3.425}, {'end': 1007.199, 'text': "Actually, a lot of students who would you know as they went through these have said they've liked doing them together.", 'start': 1001.211, 'duration': 5.988}], 'summary': 'Consider taking the practical machine learning course before this one, as many students found it helpful to do both together.', 'duration': 29.63, 'max_score': 977.569, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/IPBSB1HLNLo/pics/IPBSB1HLNLo977569.jpg'}, {'end': 1076.135, 'src': 'embed', 'start': 1043.267, 'weight': 2, 'content': [{'end': 1046.385, 'text': 'and And I have a whole blog post on this topic as well.', 'start': 1043.267, 'duration': 3.118}, {'end': 1049.687, 'text': "Okay, and we'll make sure that they're linked to that from course.fast.ai.", 'start': 1046.464, 'duration': 3.223}, {'end': 1057.452, 'text': 'And I also just wanted to say, in general with fast.ai, our philosophy is to kind of learn things on an as-needed basis.', 'start': 1049.747, 'duration': 7.705}, {'end': 1058.813, 'text': 'Yeah, exactly.', 'start': 1057.593, 'duration': 1.22}, {'end': 1061.775, 'text': "Don't try and learn everything that you think you might need first,", 'start': 1058.873, 'duration': 2.902}, {'end': 1064.497, 'text': "otherwise you'll never get around to learning the stuff you actually want to learn.", 'start': 1061.775, 'duration': 2.722}, {'end': 1069.08, 'text': 'Exactly And that shows up in deep learning, I think, particularly a lot.', 'start': 1064.958, 'duration': 4.122}, {'end': 1076.135, 'text': "Yes, Okay, so in our validation folder There's a cat's folder and a dog's folder,", 'start': 1069.281, 'duration': 6.854}], 'summary': 'Fast.ai philosophy: learn as needed to prevent delaying learning what you want.', 'duration': 32.868, 'max_score': 1043.267, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/IPBSB1HLNLo/pics/IPBSB1HLNLo1043267.jpg'}, {'end': 1200.563, 'src': 'heatmap', 'start': 1081.819, 'weight': 0.703, 'content': [{'end': 1091.626, 'text': "The reason that it's set up like this is that this is kind of the most common standard approach for how image classification data sets are shared and provided,", 'start': 1081.819, 'duration': 9.807}, {'end': 1095.795, 'text': 'and the idea is that each folder is, Tells you, the label.', 'start': 1091.626, 'duration': 4.169}, {'end': 1103.501, 'text': "so there's each of these Images is labeled cats and each of the images in the dogs folder is labeled dogs.", 'start': 1095.795, 'duration': 7.706}, {'end': 1108.985, 'text': 'Okay, this is how care us works as well, for example.', 'start': 1103.501, 'duration': 5.484}, {'end': 1117.372, 'text': 'So this is a pretty standard way to share image classification files.', 'start': 1108.985, 'duration': 8.387}, {'end': 1119.194, 'text': 'So we can have a look.', 'start': 1117.372, 'duration': 1.822}, {'end': 1126.427, 'text': 'so if you go plot dot, I am show we can see an example of a, the first of the cats.', 'start': 1119.194, 'duration': 7.233}, {'end': 1133.611, 'text': "If you haven't seen This before, this is a python 3.6 format string, so you can google for that.", 'start': 1126.427, 'duration': 7.184}, {'end': 1138.975, 'text': "if you haven't seen it, It's a very convenient way to do string formatting and we use it a lot.", 'start': 1133.611, 'duration': 5.364}, {'end': 1146.159, 'text': "So there's our cat, but we're going to mainly be interested in the underlying data that makes up that cat.", 'start': 1138.975, 'duration': 7.184}, {'end': 1150.958, 'text': "so specifically, It's an image shape.", 'start': 1146.159, 'duration': 4.799}, {'end': 1154.121, 'text': 'that is, the dimensions of the array is 198 by 179 by 3.', 'start': 1150.958, 'duration': 3.163}, {'end': 1159.667, 'text': "so it's a three-dimensional array, also called a rank 3 tensor.", 'start': 1154.121, 'duration': 5.546}, {'end': 1165.192, 'text': 'And here are the first four rows and four columns of that image.', 'start': 1159.667, 'duration': 5.525}, {'end': 1177.836, 'text': 'So, as you can see, each of those cells has three Items in it, and this is the red And blue pixel values between nought and 255.', 'start': 1165.192, 'duration': 12.644}, {'end': 1183.718, 'text': "so here's a little subset of what a picture actually looks like inside your computer.", 'start': 1177.836, 'duration': 5.882}, {'end': 1185.939, 'text': "so that's that that's will be.", 'start': 1183.718, 'duration': 2.221}, {'end': 1194.621, 'text': 'our idea is to take these kinds of numbers and Use them to predict whether those kinds of numbers represent a cat or a dog,', 'start': 1185.939, 'duration': 8.682}, {'end': 1197.322, 'text': 'Based on looking at lots of pictures of cats and dogs.', 'start': 1194.621, 'duration': 2.701}, {'end': 1200.563, 'text': "so that's a pretty hard thing to do.", 'start': 1197.322, 'duration': 3.241}], 'summary': 'Standard approach for sharing image classification data sets, using three-dimensional arrays and python 3.6 format string.', 'duration': 118.744, 'max_score': 1081.819, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/IPBSB1HLNLo/pics/IPBSB1HLNLo1081819.jpg'}, {'end': 1328.853, 'src': 'embed', 'start': 1299.775, 'weight': 0, 'content': [{'end': 1303.837, 'text': 'This actually would have won the Kaggle competition of that time.', 'start': 1299.775, 'duration': 4.062}, {'end': 1309.398, 'text': "the best in the Kaggle competition was 98.9, And we're getting about 99%.", 'start': 1303.837, 'duration': 5.561}, {'end': 1320.028, 'text': "so this may surprise you that we're getting a Kaggle winning as of 20 end of 2012, early 2013.", 'start': 1309.398, 'duration': 10.63}, {'end': 1328.853, 'text': 'Kaggle winning image classifier in 17 seconds, But and three lines of code.', 'start': 1320.028, 'duration': 8.825}], 'summary': "Achieved 99% accuracy, surpassing kaggle's 98.9% record, potentially winning the competition in 17 seconds with three lines of code.", 'duration': 29.078, 'max_score': 1299.775, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/IPBSB1HLNLo/pics/IPBSB1HLNLo1299775.jpg'}, {'end': 1378.326, 'src': 'embed', 'start': 1355.313, 'weight': 1, 'content': [{'end': 1363.621, 'text': 'The FastAI library is basically a library which takes all of the best practices, approaches that we can find,', 'start': 1355.313, 'duration': 8.308}, {'end': 1367.385, 'text': 'and so each time a paper comes out that looks interesting,', 'start': 1363.621, 'duration': 3.764}, {'end': 1372.482, 'text': 'we test it If it works well for a variety of data sets and we can figure out how to tune it.', 'start': 1367.385, 'duration': 5.097}, {'end': 1378.326, 'text': 'We implement it in fast AI and so fast I kind of curates all this stuff and packages up for you,', 'start': 1372.482, 'duration': 5.844}], 'summary': 'Fastai library curates and implements best practices from new papers for a variety of datasets.', 'duration': 23.013, 'max_score': 1355.313, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/IPBSB1HLNLo/pics/IPBSB1HLNLo1355313.jpg'}], 'start': 926.191, 'title': 'Image classification and model training', 'summary': 'Discusses setting up image classification data sets, using bash commands and python variables, and the philosophy of learning on an as-needed basis. it also demonstrates training a model using fastai library, achieving about 99% accuracy in 17 seconds and emphasizes the simplicity and flexibility of the fastai library.', 'chapters': [{'end': 1221.623, 'start': 926.191, 'title': 'Image classification and machine learning basics', 'summary': 'Discusses setting up a path for image classification data sets, using bash commands and python variables, and the standard approach for sharing image classification files. it also emphasizes the importance of understanding training and validation sets in machine learning, and the philosophy of learning on an as-needed basis.', 'duration': 295.432, 'highlights': ['Setting up a path for image classification data sets and using bash commands with Python variables The speaker discusses setting up a path to data slash dogs cats and using an exclamation mark to run bash commands that refer to Python variables.', 'Importance of understanding training and validation sets in machine learning The speaker emphasizes the significance of understanding training and validation sets in machine learning, suggesting that it would be a good idea to check out a practical machine learning course for the basics.', 'Standard approach for sharing image classification files The most common standard approach for sharing image classification data sets is discussed, where each folder represents a label, as in the case of cats and dogs folders.', 'Philosophy of learning on an as-needed basis The philosophy of fast.ai is to learn things on an as-needed basis rather than trying to learn everything that might be needed first, with an emphasis on practical learning.']}, {'end': 1576.894, 'start': 1221.623, 'title': 'Training a model with fastai library', 'summary': 'Demonstrates training a model using fastai library, achieving about 99% accuracy in 17 seconds, which would have won a kaggle competition in 2012, and emphasizes the simplicity and flexibility of the fastai library and its foundation on pytorch for deep learning and machine learning.', 'duration': 355.271, 'highlights': ['The model achieved about 99% accuracy in 17 seconds, which would have won the Kaggle competition in 2012 with the best in the competition being 98.9%. Demonstrates the impressive performance of the model and its competitive edge in comparison to the Kaggle competition.', 'The FastAI library is built on top of best practices and approaches, simplifying the process of training the model and automatically figuring out the best way to handle things. Highlights the simplicity and flexibility of the FastAI library, which automates the handling of best practices and approaches, making the training process easier.', 'FastAI library is based on PyTorch, a flexible deep learning and machine learning library favored by top researchers. Emphasizes the foundation of the FastAI library on PyTorch, a flexible and favored library among top researchers.']}], 'duration': 650.703, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/IPBSB1HLNLo/pics/IPBSB1HLNLo926191.jpg', 'highlights': ['The model achieved about 99% accuracy in 17 seconds, outperforming the best in the Kaggle competition.', 'The FastAI library simplifies the training process and automatically figures out the best way to handle things.', 'The philosophy of fast.ai is to learn things on an as-needed basis, emphasizing practical learning.', 'The FastAI library is based on PyTorch, a flexible deep learning and machine learning library favored by top researchers.', 'Importance of understanding training and validation sets in machine learning.']}, {'end': 2283.922, 'segs': [{'end': 1631.785, 'src': 'embed', 'start': 1576.894, 'weight': 3, 'content': [{'end': 1579.395, 'text': 'and so here are some images that it was correct about.', 'start': 1576.894, 'duration': 2.501}, {'end': 1586.517, 'text': 'okay, and so remember, one is a dog, so anything greater than 0.5 is dog and Zero is a cat.', 'start': 1579.395, 'duration': 7.122}, {'end': 1590.659, 'text': 'So this is what 10 to the negative 5, obviously a cat.', 'start': 1586.758, 'duration': 3.901}, {'end': 1593.82, 'text': 'Here are some which are incorrect.', 'start': 1590.659, 'duration': 3.161}, {'end': 1601.862, 'text': "right. so you can see that some of these which it thinks are incorrect, obviously are just The you know images that shouldn't be there at all.", 'start': 1593.82, 'duration': 8.042}, {'end': 1606.843, 'text': 'but clearly this one, which it called a Dog, is not at all a dog.', 'start': 1601.862, 'duration': 4.981}, {'end': 1611.627, 'text': 'so there are some obvious mistakes.', 'start': 1606.843, 'duration': 4.784}, {'end': 1617.532, 'text': 'We can also take a look at Which cats is it the most confident?', 'start': 1611.627, 'duration': 5.905}, {'end': 1617.993, 'text': 'are cats?', 'start': 1617.532, 'duration': 0.461}, {'end': 1622.517, 'text': 'Which dogs are the most dog like the most confident dogs?', 'start': 1619.054, 'duration': 3.463}, {'end': 1629.423, 'text': 'Perhaps, more interestingly, we can also see which cats is it the most confident are actually dogs.', 'start': 1624.158, 'duration': 5.265}, {'end': 1631.785, 'text': 'so which ones it is at the most wrong about.', 'start': 1629.423, 'duration': 2.362}], 'summary': 'Analyzing image recognition accuracy and identifying obvious mistakes and misclassifications.', 'duration': 54.891, 'max_score': 1576.894, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/IPBSB1HLNLo/pics/IPBSB1HLNLo1576894.jpg'}, {'end': 1736.517, 'src': 'embed', 'start': 1678.04, 'weight': 1, 'content': [{'end': 1684.664, 'text': "Because if I want to make the model better, Then I need to take advantage of the things it's doing well and fix the things it's doing badly.", 'start': 1678.04, 'duration': 6.624}, {'end': 1691.348, 'text': "So in this case And often this is the case I've learned something about the data set itself,", 'start': 1685.405, 'duration': 5.943}, {'end': 1697.752, 'text': "Which is that there are some things that are in here that probably shouldn't be, But I've also like.", 'start': 1691.348, 'duration': 6.404}, {'end': 1703.227, 'text': "it's also clear that this model has room to improve.", 'start': 1697.752, 'duration': 5.475}, {'end': 1706.969, 'text': "To me, that's pretty obviously a dog.", 'start': 1704.307, 'duration': 2.662}, {'end': 1715.693, 'text': "But one thing I'm suspicious about here is this image is very kind of fat and short.", 'start': 1707.609, 'duration': 8.084}, {'end': 1724.407, 'text': 'as we all learn The way these algorithms work, is it kind of grabs a square piece at a time?', 'start': 1717.501, 'duration': 6.906}, {'end': 1732.393, 'text': "So this rather makes me suspicious that we're going to need to use something called data augmentation, that we'll learn about later,", 'start': 1725.528, 'duration': 6.865}, {'end': 1736.517, 'text': 'to handle this properly.', 'start': 1732.393, 'duration': 4.124}], 'summary': 'Model needs improvement, data may have irrelevant elements, need for data augmentation.', 'duration': 58.477, 'max_score': 1678.04, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/IPBSB1HLNLo/pics/IPBSB1HLNLo1678040.jpg'}, {'end': 1823.33, 'src': 'embed', 'start': 1795.805, 'weight': 0, 'content': [{'end': 1805.727, 'text': 'you can run exactly the same three lines of code and just point your path variable somewhere else To get your own image classifier.', 'start': 1795.805, 'duration': 9.922}, {'end': 1813.388, 'text': 'so, for example, one student Took those three lines of code downloaded from Google images.', 'start': 1805.727, 'duration': 7.661}, {'end': 1821.83, 'text': 'ten examples of pictures of people playing cricket, ten examples of people playing baseball and build a classifier of those images,', 'start': 1813.388, 'duration': 8.442}, {'end': 1823.33, 'text': 'which was nearly perfectly correct.', 'start': 1821.83, 'duration': 1.5}], 'summary': 'Using three lines of code, a student built a nearly perfect image classifier with 10 examples of people playing cricket and 10 examples of people playing baseball.', 'duration': 27.525, 'max_score': 1795.805, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/IPBSB1HLNLo/pics/IPBSB1HLNLo1795805.jpg'}], 'start': 1576.894, 'title': 'Visualizing image classifier model and top-down learning approach', 'summary': "Discusses visualizing an image classifier model, analyzing its mistakes, identifying confident predictions, and the need for data augmentation and diverse image sets. it also introduces a top-down learning approach, gradually delving into neural network training, problem-solving, and library usage, and culminating in the creation of world-class image, structured data, language, and recommendation system classifiers. the model's performance is demonstrated with examples of cricket and currency images, achieving nearly perfect accuracy.", 'chapters': [{'end': 1848.347, 'start': 1576.894, 'title': 'Visualizing image classifier model', 'summary': "Discusses visualizing an image classifier model, analyzing its mistakes, identifying confident predictions, and the need for data augmentation and diverse image sets. the model's performance is demonstrated with examples of cricket and currency images, achieving nearly perfect accuracy.", 'duration': 271.453, 'highlights': ["The model's mistakes are evident in misclassified images, including a dog misidentified as a cat, indicating room for improvement.", "Analyzing the model's confidence in predictions reveals the most confident cats and dogs, as well as instances where cats are mistaken for dogs and vice versa.", "The importance of visualizing the model's output to understand its strengths and weaknesses in order to improve it and gain insights into the dataset.", "The need for data augmentation is highlighted to handle images that may not fit the model's current criteria, such as short and fat images.", "Demonstration of the model's accuracy with examples of cricket and currency images, achieving nearly perfect results and encouraging experimentation with diverse image sets."]}, {'end': 2283.922, 'start': 1848.347, 'title': 'Top-down learning approach', 'summary': 'Introduces a top-down learning approach, gradually delving into neural network training, problem-solving, and library usage, and culminating in the creation of world-class image, structured data, language, and recommendation system classifiers.', 'duration': 435.575, 'highlights': ['The chapter introduces a top-down learning approach, gradually delving into neural network training, problem-solving, and library usage. The top-down learning approach aims to teach problem-solving by gradually delving into neural network training and library usage.', 'The course aims to teach the creation of world-class image, structured data, language, and recommendation system classifiers. By the end of the course, students will learn to create world-class image, structured data, language, and recommendation system classifiers.', 'Students are encouraged to experiment and delve into the code-driven approach, peeling back the layers to understand the details. The course encourages students to experiment and explore the code-driven approach, emphasizing understanding the details by peeling back the layers.', 'The learning process involves watching the videos multiple times, with an emphasis on completing the entire course before aiming to fully understand the details. Students are advised to watch the videos multiple times and complete the entire course before aiming to fully understand the details, promoting a comprehensive learning approach.']}], 'duration': 707.028, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/IPBSB1HLNLo/pics/IPBSB1HLNLo1576894.jpg', 'highlights': ["Demonstration of the model's accuracy with examples of cricket and currency images, achieving nearly perfect results and encouraging experimentation with diverse image sets.", "The importance of visualizing the model's output to understand its strengths and weaknesses in order to improve it and gain insights into the dataset.", "The need for data augmentation is highlighted to handle images that may not fit the model's current criteria, such as short and fat images.", "The model's mistakes are evident in misclassified images, including a dog misidentified as a cat, indicating room for improvement.", "Analyzing the model's confidence in predictions reveals the most confident cats and dogs, as well as instances where cats are mistaken for dogs and vice versa."]}, {'end': 3128.786, 'segs': [{'end': 2309.023, 'src': 'embed', 'start': 2286.192, 'weight': 0, 'content': [{'end': 2294.636, 'text': "From there we're going to then dig back into language a bit more and we're going to look at Actually we're going to look at the writings of Nietzsche,", 'start': 2286.192, 'duration': 8.444}, {'end': 2301.679, 'text': 'the philosopher, and learn how to create our own Nietzsche philosophy from scratch, Character by character.', 'start': 2294.636, 'duration': 7.043}, {'end': 2309.023, 'text': 'so this here, perhaps, that every life values of blood, of intercourse, when it senses there is unscrupulous, his very rights and still impulse,', 'start': 2301.679, 'duration': 7.344}], 'summary': "Analyzing nietzsche's writings to create a personal philosophy, character by character.", 'duration': 22.831, 'max_score': 2286.192, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/IPBSB1HLNLo/pics/IPBSB1HLNLo2286192.jpg'}, {'end': 2502.325, 'src': 'embed', 'start': 2477.361, 'weight': 2, 'content': [{'end': 2483.851, 'text': 'you know, Have created patents based on deep learning and so forth, who have done it by doing this course.', 'start': 2477.361, 'duration': 6.49}, {'end': 2486.416, 'text': 'So the top-down approach works super well.', 'start': 2483.851, 'duration': 2.565}, {'end': 2498.202, 'text': "Now one thing to mention is like we've, we've now already learned how you can actually train a world-class image classifier in 17 seconds.", 'start': 2487.935, 'duration': 10.267}, {'end': 2502.325, 'text': 'I should mention, by the way, the first time you run that code.', 'start': 2498.202, 'duration': 4.123}], 'summary': 'Patented deep learning technology, achieving world-class image classifier in 17 seconds.', 'duration': 24.964, 'max_score': 2477.361, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/IPBSB1HLNLo/pics/IPBSB1HLNLo2477361.jpg'}, {'end': 2589.481, 'src': 'embed', 'start': 2554.167, 'weight': 3, 'content': [{'end': 2560.111, 'text': 'Image classification algorithms are really useful for lots and lots of things.', 'start': 2554.167, 'duration': 5.944}, {'end': 2565.355, 'text': 'for example, Alpha go, which became which beat the go world champion?', 'start': 2560.111, 'duration': 5.244}, {'end': 2577.319, 'text': 'The way it worked was to use something At its heart that looked almost exactly like our dogs versus cats image classification algorithm.', 'start': 2566.056, 'duration': 11.263}, {'end': 2589.481, 'text': 'It looked at thousands and thousands of go boards And at for each one there was a label saying whether that go board ended up being the winning or the losing player,', 'start': 2577.319, 'duration': 12.162}], 'summary': 'Image classification algorithms, like alphago, analyzed thousands of go boards to predict winning or losing outcomes.', 'duration': 35.314, 'max_score': 2554.167, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/IPBSB1HLNLo/pics/IPBSB1HLNLo2554167.jpg'}], 'start': 2286.192, 'title': 'Using neural network for philosophy and image recognition', 'summary': 'Delves into creating nietzsche philosophy character by character using recurrent neural network and discusses the practical applications of image recognition models, emphasizing the speed improvement in training world-class image classifiers using gpus.', 'chapters': [{'end': 2324.487, 'start': 2286.192, 'title': 'Creating nietzsche philosophy through neural network', 'summary': 'Explores creating nietzsche philosophy character by character using recurrent neural network, with a focus on language and computer vision.', 'duration': 38.295, 'highlights': ['The chapter delves into creating Nietzsche philosophy from scratch character by character using a recurrent neural network.', 'It emphasizes the exploration of language and the application of computer vision in the process.', "The focus is on understanding Nietzsche's writings and generating text character by character."]}, {'end': 3128.786, 'start': 2324.487, 'title': 'Deep learning for image recognition', 'summary': 'Covers the importance of hands-on coding, building image recognition models, and the practical applications of image classifiers in various fields. it also emphasizes the effectiveness of a top-down approach to learning deep learning, and highlights the speed improvement in training world-class image classifiers using gpus.', 'duration': 804.299, 'highlights': ['Speed improvement in training image classifiers using GPUs Training world-class image classifiers in 17 seconds with GPU, compared to several minutes on CPU, emphasizes the significant speed enhancement for deep learning tasks.', 'Practical applications of image classifiers Examples of using image classifiers beyond basic image recognition, such as in AlphaGo for analyzing Go boards and in anti-fraud measures based on mouse movement patterns, highlight the diverse and practical applications of deep learning.', 'Top-down approach to learning deep learning The effectiveness of a practical, code-first approach to learning deep learning, as opposed to a theory-heavy approach, is highlighted through successful outcomes of thousands of course participants in running research labs and creating patents based on deep learning.', 'Importance of hands-on coding The significance of spending more time on running code and practical implementation, rather than theoretical knowledge, is emphasized to effectively become an ML practitioner.']}], 'duration': 842.594, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/IPBSB1HLNLo/pics/IPBSB1HLNLo2286192.jpg', 'highlights': ['The chapter delves into creating Nietzsche philosophy from scratch character by character using a recurrent neural network.', "The focus is on understanding Nietzsche's writings and generating text character by character.", 'The chapter emphasizes the exploration of language and the application of computer vision in the process.', 'Practical applications of image classifiers beyond basic image recognition are highlighted, such as in AlphaGo for analyzing Go boards and in anti-fraud measures based on mouse movement patterns.', 'The speed improvement in training world-class image classifiers using GPUs is emphasized, with training taking 17 seconds with GPU compared to several minutes on CPU.', 'The top-down approach to learning deep learning and the importance of hands-on coding for becoming an ML practitioner are emphasized.']}, {'end': 3561.797, 'segs': [{'end': 3157.023, 'src': 'embed', 'start': 3128.786, 'weight': 0, 'content': [{'end': 3140.993, 'text': 'So GPUs turn out to be able to solve these neural network parameter fitting problems Incredibly quickly and also incredibly cheaply.', 'start': 3128.786, 'duration': 12.207}, {'end': 3148.176, 'text': "so they've been absolutely key in bringing these three pieces together.", 'start': 3140.993, 'duration': 7.183}, {'end': 3152.439, 'text': "Then there's one more piece, Which is, I mentioned that these neural networks.", 'start': 3148.176, 'duration': 4.263}, {'end': 3157.023, 'text': 'you can intersperse multiple sets of linear and then nonlinear layers.', 'start': 3152.439, 'duration': 4.584}], 'summary': 'Gpus are instrumental in quickly and cost-effectively solving neural network parameter fitting problems, a key factor in combining neural networks with linear and nonlinear layers.', 'duration': 28.237, 'max_score': 3128.786, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/IPBSB1HLNLo/pics/IPBSB1HLNLo3128786.jpg'}, {'end': 3209.011, 'src': 'embed', 'start': 3180.479, 'weight': 1, 'content': [{'end': 3185.601, 'text': 'They require an exponentially increasing number of parameters to do so,', 'start': 3180.479, 'duration': 5.122}, {'end': 3191.724, 'text': "so they don't actually solve the fast and scalable for even reasonable size problems.", 'start': 3185.601, 'duration': 6.123}, {'end': 3200.007, 'text': "But we've since discovered that if you create, add multiple hidden layers, Then you get super linear scaling,", 'start': 3191.724, 'duration': 8.283}, {'end': 3209.011, 'text': 'so you can add a few more hidden layers to get multiplicatively more accuracy to multiplicatively more complex problems.', 'start': 3200.007, 'duration': 9.004}], 'summary': 'Adding multiple hidden layers leads to super linear scaling for accuracy and complexity.', 'duration': 28.532, 'max_score': 3180.479, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/IPBSB1HLNLo/pics/IPBSB1HLNLo3180479.jpg'}, {'end': 3345.288, 'src': 'embed', 'start': 3318.816, 'weight': 3, 'content': [{'end': 3328.002, 'text': "And so it's actually using deep learning here to read the original email and to generate some suggested replies and so like.", 'start': 3318.816, 'duration': 9.186}, {'end': 3334.385, 'text': "this is a really great example of the kind of stuff that previously just wasn't possible.", 'start': 3328.002, 'duration': 6.383}, {'end': 3345.288, 'text': 'Another great example would be Microsoft is also a little bit more recently invested heavily in deep learning, and so now you can use Skype.', 'start': 3334.385, 'duration': 10.903}], 'summary': 'Deep learning enables email response generation, making impossible tasks possible. microsoft heavily invested in deep learning for skype.', 'duration': 26.472, 'max_score': 3318.816, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/IPBSB1HLNLo/pics/IPBSB1HLNLo3318816.jpg'}, {'end': 3511.268, 'src': 'embed', 'start': 3484.348, 'weight': 4, 'content': [{'end': 3494.595, 'text': "and since that time the idea of using deep learning for medical imaging has become Hugely popular and it's being used all around the world.", 'start': 3484.348, 'duration': 10.247}, {'end': 3505.945, 'text': "So what I've generally noticed is that you know the vast majority of Have kind of That people do in the world currently aren't using deep learning,", 'start': 3494.595, 'duration': 11.35}, {'end': 3511.268, 'text': "and then each time somebody says, oh, Let's try using deep learning to improve performance at this thing,", 'start': 3505.945, 'duration': 5.323}], 'summary': "Deep learning for medical imaging is hugely popular worldwide, but most people aren't using it.", 'duration': 26.92, 'max_score': 3484.348, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/IPBSB1HLNLo/pics/IPBSB1HLNLo3484348.jpg'}, {'end': 3561.797, 'src': 'embed', 'start': 3537.047, 'weight': 5, 'content': [{'end': 3544.49, 'text': 'and You know things which people spend a lot of money on or have a lot of you know important business opportunities.', 'start': 3537.047, 'duration': 7.443}, {'end': 3551.433, 'text': "There's lots more as well, But these are some examples of things that maybe at your company you could think about applying deep learning for.", 'start': 3544.49, 'duration': 6.943}, {'end': 3561.797, 'text': "So let's talk about what's actually going on, What actually happened when we trained that deep learning model earlier?", 'start': 3553.689, 'duration': 8.108}], 'summary': 'Deep learning can save money and help identify business opportunities at companies.', 'duration': 24.75, 'max_score': 3537.047, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/IPBSB1HLNLo/pics/IPBSB1HLNLo3537047.jpg'}], 'start': 3128.786, 'title': 'Gpus and deep learning impact', 'summary': 'Discusses the role of gpus in accelerating neural network parameter fitting, leading to increased accuracy with multiple hidden layers, and highlights the exponential growth and impact of deep learning adoption by major companies like google and microsoft, showcasing its impact in email replies, real-time language translation in skype, and medical imaging, as well as its potential applications in various industries.', 'chapters': [{'end': 3209.011, 'start': 3128.786, 'title': 'Role of gpus in neural network parameter fitting', 'summary': 'Discusses the pivotal role of gpus in accelerating neural network parameter fitting, leading to exponentially increasing accuracy with the addition of multiple hidden layers.', 'duration': 80.225, 'highlights': ['GPUs play a key role in solving neural network parameter fitting problems incredibly quickly and cheaply. GPUs are crucial for fast and cost-effective solutions to neural network parameter fitting.', 'Addition of multiple hidden layers leads to super linear scaling, resulting in multiplicatively more accuracy for complex problems. The inclusion of multiple hidden layers results in exponentially increasing accuracy for increasingly complex problems.', 'Neural networks require an exponentially increasing number of parameters to solve problems arbitrarily closely. Neural networks need an exponentially growing number of parameters for solving problems closely.']}, {'end': 3561.797, 'start': 3210.336, 'title': 'Deep learning: impact and applications', 'summary': 'Highlights the exponential growth of deep learning adoption by major companies like google and microsoft, showcasing its impact in email replies, real-time language translation in skype, and medical imaging. it also emphasizes the potential applications of deep learning in various industries, leading to substantial improvements in performance and business opportunities.', 'duration': 351.461, 'highlights': ["Deep learning's exponential growth in adoption by major companies like Google and Microsoft, demonstrated through examples such as email replies, real-time language translation in Skype, and medical imaging. The exponential growth of deep learning's usage at Google and Microsoft, with applications in email replies, real-time language translation, and medical imaging, showcases its impact and potential applications across different industries.", "Demonstration of deep learning's impact in improving medical imaging accuracy, leading to the successful establishment of a startup called analytic, which has gained significant popularity globally. The successful use of deep learning in medical imaging, leading to the establishment of the startup analytic, demonstrates its potential in significantly improving accuracy and gaining global popularity in the medical field.", 'Exploration of potential applications for deep learning in various industries, with specific examples provided, emphasizing the substantial improvements in performance and business opportunities. The exploration of potential applications for deep learning in various industries, with specific examples provided, emphasizes the substantial improvements in performance and business opportunities, showcasing the wide-ranging potential of deep learning.']}], 'duration': 433.011, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/IPBSB1HLNLo/pics/IPBSB1HLNLo3128786.jpg', 'highlights': ['GPUs play a key role in solving neural network parameter fitting problems incredibly quickly and cheaply.', 'Addition of multiple hidden layers leads to super linear scaling, resulting in multiplicatively more accuracy for complex problems.', 'Neural networks require an exponentially increasing number of parameters to solve problems arbitrarily closely.', "Deep learning's exponential growth in adoption by major companies like Google and Microsoft, demonstrated through examples such as email replies, real-time language translation in Skype, and medical imaging.", "Demonstration of deep learning's impact in improving medical imaging accuracy, leading to the successful establishment of a startup called analytic, which has gained significant popularity globally.", 'Exploration of potential applications for deep learning in various industries, with specific examples provided, emphasizing the substantial improvements in performance and business opportunities.']}, {'end': 4058.548, 'segs': [{'end': 3592.426, 'src': 'embed', 'start': 3562.758, 'weight': 0, 'content': [{'end': 3570.365, 'text': 'and so, as I briefly mentioned, the thing we created is something called a convolutional neural network, or CNN,', 'start': 3562.758, 'duration': 7.607}, {'end': 3574.929, 'text': 'and The key piece of a convolutional neural network is the convolution.', 'start': 3570.365, 'duration': 4.564}, {'end': 3579.519, 'text': "So here's a great example from a website.", 'start': 3576.417, 'duration': 3.102}, {'end': 3584.582, 'text': "I've got the URL up here, explained visually.", 'start': 3579.519, 'duration': 5.063}, {'end': 3592.426, 'text': "It's called, and the explained visually website has an example of a convolution Kind of in practice.", 'start': 3584.582, 'duration': 7.844}], 'summary': 'Created a convolutional neural network (cnn) with a key piece being the convolution.', 'duration': 29.668, 'max_score': 3562.758, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/IPBSB1HLNLo/pics/IPBSB1HLNLo3562758.jpg'}, {'end': 3748.097, 'src': 'heatmap', 'start': 3681.83, 'weight': 0.726, 'content': [{'end': 3689.076, 'text': 'And so if you do that for every 3x3 area, you end up with the values that you see over here on the right-hand side.', 'start': 3681.83, 'duration': 7.246}, {'end': 3696.922, 'text': 'So very low values become black, very high values become white,', 'start': 3690.897, 'duration': 6.025}, {'end': 3704.368, 'text': "and so you can see when we're at an edge where it's black at the bottom and white at the top.", 'start': 3696.922, 'duration': 7.446}, {'end': 3706.89, 'text': "we're obviously going to get higher numbers over here.", 'start': 3704.368, 'duration': 2.522}, {'end': 3708.554, 'text': 'and vice versa.', 'start': 3707.754, 'duration': 0.8}, {'end': 3711.075, 'text': "So that's a convolution.", 'start': 3709.895, 'duration': 1.18}, {'end': 3718.979, 'text': 'So, as you can see, it is a linear operation and so, based on that definition of a neural net I described before,', 'start': 3712.276, 'duration': 6.703}, {'end': 3721.56, 'text': 'this can be a layer in our neural network.', 'start': 3718.979, 'duration': 2.581}, {'end': 3724.241, 'text': 'It is a simple linear operation.', 'start': 3722.06, 'duration': 2.181}, {'end': 3731.164, 'text': "And we're going to look much more at convolutions later, including building a little spreadsheet that implements them ourselves.", 'start': 3725.041, 'duration': 6.123}, {'end': 3736.346, 'text': "So the next thing we're going to do is we're going to add a nonlinear layer.", 'start': 3732.484, 'duration': 3.862}, {'end': 3748.097, 'text': "So a non-linearity, as it's called, is something which takes an input value and turns it into some different value in a non-linear way.", 'start': 3737.594, 'duration': 10.503}], 'summary': 'Convolution operation can be a layer in a neural network, adding non-linear layer next.', 'duration': 66.267, 'max_score': 3681.83, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/IPBSB1HLNLo/pics/IPBSB1HLNLo3681830.jpg'}, {'end': 3865.278, 'src': 'embed', 'start': 3837.399, 'weight': 1, 'content': [{'end': 3843.483, 'text': 'build these arbitrarily tall or thin blocks and then combine those blocks together.', 'start': 3837.399, 'duration': 6.084}, {'end': 3852.929, 'text': 'and this is actually the essence of the universal approximation theorem, this idea that when you have a linear layer of Feeding into a non linearity,', 'start': 3843.483, 'duration': 9.446}, {'end': 3856.551, 'text': 'you can actually create these arbitrarily complex shapes.', 'start': 3852.929, 'duration': 3.622}, {'end': 3865.278, 'text': 'So this is the key idea behind why neural networks can solve any computable problem.', 'start': 3856.551, 'duration': 8.727}], 'summary': 'Neural networks can create complex shapes, solving any computable problem.', 'duration': 27.879, 'max_score': 3837.399, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/IPBSB1HLNLo/pics/IPBSB1HLNLo3837399.jpg'}, {'end': 3924.098, 'src': 'embed', 'start': 3889.304, 'weight': 2, 'content': [{'end': 3895.085, 'text': 'This is an extract from a notebook, actually one of the fast AI lessons,', 'start': 3889.304, 'duration': 5.781}, {'end': 3901.507, 'text': 'And it shows actually an example of using gradient descent to solve a simple linear regression problem.', 'start': 3895.085, 'duration': 6.422}, {'end': 3904.831, 'text': 'But I can show you the basic idea.', 'start': 3903.37, 'duration': 1.461}, {'end': 3918.036, 'text': "Let's say you had a simple quadratic, and so you were trying to find the minimum of this quadratic.", 'start': 3905.531, 'duration': 12.505}, {'end': 3924.098, 'text': 'And so in order to find the minimum, you start out by randomly picking some point.', 'start': 3918.996, 'duration': 5.102}], 'summary': 'An example of using gradient descent to solve a simple linear regression problem.', 'duration': 34.794, 'max_score': 3889.304, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/IPBSB1HLNLo/pics/IPBSB1HLNLo3889304.jpg'}], 'start': 3562.758, 'title': 'Neural networks and cnn', 'summary': 'Covers the concepts of convolutional neural network (cnn) and its application in edge detection and image transformation. it also explores linear operations, nonlinearities, sigmoid and relu functions, the universal approximation theorem, and the use of gradient descent for parameter optimization.', 'chapters': [{'end': 3711.075, 'start': 3562.758, 'title': 'Convolutional neural network', 'summary': 'Discusses the concept of convolutional neural network (cnn) and illustrates the process of convolution using a 3x3 kernel, resulting in edge detection and image transformation.', 'duration': 148.317, 'highlights': ['Convolutional neural network (CNN) is explained using the concept of convolution and a 3x3 kernel, which results in edge detection and image transformation.', 'The specific 3x3 kernel values used in the example are 1, 2, 1, 0, -1, -2, -1, which are multiplied with the corresponding pixel values to obtain the transformed image.', 'The process of convolution involves multiplying the pixel values of each 3x3 area with the kernel values and adding them together to obtain the transformed image.']}, {'end': 4058.548, 'start': 3712.276, 'title': 'Neural networks and gradient descent', 'summary': 'Explains the concept of linear operations and nonlinearities in neural networks, with a focus on the use of sigmoid and relu functions. it also delves into the universal approximation theorem and the application of gradient descent to update parameters for optimization.', 'duration': 346.272, 'highlights': ['Neural networks create complex shapes with linear layers and nonlinear functions. The combination of a linear layer followed by an element-wise nonlinear function allows the creation of arbitrarily complex shapes, as described by the universal approximation theorem.', 'Explanation of sigmoid and ReLU as popular non-linear functions in neural networks. The chapter discusses the use of sigmoid and ReLU functions as popular non-linearities in neural networks, with ReLU replacing negative numbers with 0 and leaving positive numbers unchanged.', 'Application of gradient descent for parameter optimization in neural networks. The concept of using gradient descent to update parameters for optimizing neural network models is detailed, with emphasis on the importance of choosing an appropriate step size to avoid divergence.']}], 'duration': 495.79, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/IPBSB1HLNLo/pics/IPBSB1HLNLo3562758.jpg', 'highlights': ['Convolutional neural network (CNN) explained using the concept of convolution and a 3x3 kernel for edge detection and image transformation.', 'Neural networks create complex shapes with linear layers and nonlinear functions, as described by the universal approximation theorem.', 'Application of gradient descent for parameter optimization in neural networks, emphasizing the importance of choosing an appropriate step size to avoid divergence.']}, {'end': 5193.868, 'segs': [{'end': 4141.273, 'src': 'embed', 'start': 4114.464, 'weight': 3, 'content': [{'end': 4123.45, 'text': 'and what they did a few years ago was they figured out how to basically draw a picture of what each layer in a deep learning network learned.', 'start': 4114.464, 'duration': 8.986}, {'end': 4133.996, 'text': 'And so they showed that Layer 1 of the network here are nine examples of convolutional filters from Layer 1 of a trained network.', 'start': 4124.951, 'duration': 9.045}, {'end': 4141.273, 'text': 'they found that some of the filters kind of learnt these diagonal lines or Simple little grid patterns.', 'start': 4135.269, 'duration': 6.004}], 'summary': 'Deep learning network analyzed to visualize learned patterns and filters.', 'duration': 26.809, 'max_score': 4114.464, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/IPBSB1HLNLo/pics/IPBSB1HLNLo4114464.jpg'}, {'end': 4282.214, 'src': 'embed', 'start': 4203.596, 'weight': 0, 'content': [{'end': 4212.282, 'text': "basically Something that's being activated nearly entirely by little bits of sunset, something that's being activated by circular objects,", 'start': 4203.596, 'duration': 8.686}, {'end': 4221.069, 'text': "something that's being activated by Repeating horizontal lines, something that's being activated by corners right?", 'start': 4213.203, 'duration': 7.866}, {'end': 4226.213, 'text': "So you can see how we're basically combining layer one features together.", 'start': 4221.289, 'duration': 4.924}, {'end': 4234.619, 'text': 'So if we combine those features together and again these are all convolutional filters learnt through gradient descent By the third layer,', 'start': 4226.213, 'duration': 8.406}, {'end': 4239.343, 'text': "it's actually learnt to recognize the presence of text.", 'start': 4234.619, 'duration': 4.724}, {'end': 4247.191, 'text': 'another filter has learnt to recognize the presence of petals and Another filter has learnt to recognize the presence of human faces.', 'start': 4239.343, 'duration': 7.848}, {'end': 4252.214, 'text': 'So just three layers is enough to get some pretty rich behavior.', 'start': 4248.111, 'duration': 4.103}, {'end': 4255.556, 'text': 'By the time we get to layer five.', 'start': 4253.495, 'duration': 2.061}, {'end': 4263.421, 'text': "we've got something that can recognize the eyeballs of insects and birds and something that can recognize unicycle wheels.", 'start': 4255.556, 'duration': 7.865}, {'end': 4270.326, 'text': 'So this is kind of where we start with something incredibly simple.', 'start': 4264.942, 'duration': 5.384}, {'end': 4282.214, 'text': 'But if we use it as a bit a big enough scale, Thanks to the universal approximation theorem and the use of multiple hidden layers in deep learning,', 'start': 4272.388, 'duration': 9.826}], 'summary': 'Deep learning model learns to recognize text, petals, human faces, insect eyeballs, and unicycle wheels in just three to five layers.', 'duration': 78.618, 'max_score': 4203.596, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/IPBSB1HLNLo/pics/IPBSB1HLNLo4203596.jpg'}, {'end': 4449.552, 'src': 'embed', 'start': 4416.097, 'weight': 4, 'content': [{'end': 4428.544, 'text': "cyclical learning rates for training neural networks by a terrific researcher called Leslie Smith, and I'll show you Leslie's idea.", 'start': 4416.097, 'duration': 12.447}, {'end': 4438.85, 'text': "So Leslie's idea started out with the same Basic idea that we've seen before, which is, if we're going to optimize something, pick some random point,", 'start': 4428.544, 'duration': 10.306}, {'end': 4449.552, 'text': 'Take its gradient, And then specifically, he said take a tiny, tiny step, Like tiny step.', 'start': 4438.85, 'duration': 10.702}], 'summary': "Leslie smith's idea involves using cyclical learning rates for training neural networks.", 'duration': 33.455, 'max_score': 4416.097, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/IPBSB1HLNLo/pics/IPBSB1HLNLo4416097.jpg'}, {'end': 4672.694, 'src': 'embed', 'start': 4602.115, 'weight': 5, 'content': [{'end': 4613.937, 'text': "and?. So if you create your learn object in the same way that we did before, we'll be learning more about these details shortly.", 'start': 4602.115, 'duration': 11.822}, {'end': 4623.185, 'text': "If you then call LRFind method on that, you'll see that it'll start training Model like it did before,", 'start': 4614.858, 'duration': 8.327}, {'end': 4626.789, 'text': "but it'll generally stop before it gets to a hundred percent.", 'start': 4623.185, 'duration': 3.604}, {'end': 4635.238, 'text': "because if it notices That the loss is getting a lot worse, Then it'll stop automatically.", 'start': 4626.789, 'duration': 8.449}, {'end': 4636.079, 'text': 'that so that you can see here.', 'start': 4635.238, 'duration': 0.841}, {'end': 4640.916, 'text': 'It stopped at 84% and so then you can call a Learn.shed.', 'start': 4636.099, 'duration': 4.817}, {'end': 4642.897, 'text': 'that gets you the learning rate scheduler.', 'start': 4640.916, 'duration': 1.981}, {'end': 4648.701, 'text': "That's the object which actually does this learning rate finding, and that object has a plot, learning rate function.", 'start': 4642.897, 'duration': 5.804}, {'end': 4653.444, 'text': 'And so you can see here over by iteration, you can see the learning rate right?', 'start': 4648.701, 'duration': 4.743}, {'end': 4660.623, 'text': 'So you can see each step, the learning rates getting bigger and bigger and You can do it this way.', 'start': 4653.464, 'duration': 7.159}, {'end': 4662.965, 'text': "we can see it's increasing exponentially.", 'start': 4660.623, 'duration': 2.342}, {'end': 4667.989, 'text': 'another way that Leslie Smith, the researcher, suggests is to do it linearly.', 'start': 4662.965, 'duration': 5.024}, {'end': 4672.694, 'text': "So I'm actually currently researching with both of these approaches to see which works best.", 'start': 4667.989, 'duration': 4.705}], 'summary': 'Using lrfind method, training stops at 84%. researching exponential vs. linear learning rate increase.', 'duration': 70.579, 'max_score': 4602.115, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/IPBSB1HLNLo/pics/IPBSB1HLNLo4602115.jpg'}, {'end': 4806.553, 'src': 'embed', 'start': 4778.449, 'weight': 9, 'content': [{'end': 4789.761, 'text': 'so a question of How many epochs should we run is kind of the one other question that you need to answer to run these three lines of code,', 'start': 4778.449, 'duration': 11.312}, {'end': 4795.726, 'text': 'and the answer really, to me is like As many as you like.', 'start': 4789.761, 'duration': 5.965}, {'end': 4804.152, 'text': "What you might find happen is, if you run it for too long, the accuracy you'll start getting worse Right and we'll learn about that.", 'start': 4795.726, 'duration': 8.426}, {'end': 4804.712, 'text': 'why later?', 'start': 4804.152, 'duration': 0.56}, {'end': 4806.553, 'text': "It's something called overfitting, right?", 'start': 4804.792, 'duration': 1.761}], 'summary': 'Run as many epochs as desired, but watch for overfitting.', 'duration': 28.104, 'max_score': 4778.449, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/IPBSB1HLNLo/pics/IPBSB1HLNLo4778449.jpg'}, {'end': 4883.653, 'src': 'embed', 'start': 4855.087, 'weight': 10, 'content': [{'end': 4857.488, 'text': 'try to get a sense of like.', 'start': 4855.087, 'duration': 2.401}, {'end': 4859.91, 'text': 'Which kinds of images does this seem to work well for??', 'start': 4857.488, 'duration': 2.422}, {'end': 4862.932, 'text': "Which ones doesn't it work work well for?", 'start': 4861.01, 'duration': 1.922}, {'end': 4867.915, 'text': 'What kind of learning rates do you need for different kinds of images?', 'start': 4864.853, 'duration': 3.062}, {'end': 4869.236, 'text': 'How many epochs do you need??', 'start': 4868.015, 'duration': 1.221}, {'end': 4873.659, 'text': 'How does the number of the learning rate change, the accuracy you get?', 'start': 4869.976, 'duration': 3.683}, {'end': 4879.471, 'text': 'and so forth, like, really experiment and then, You know, try to get a sense of like.', 'start': 4873.659, 'duration': 5.812}, {'end': 4881.212, 'text': "what's inside this data object?", 'start': 4879.471, 'duration': 1.741}, {'end': 4883.653, 'text': 'You know what are the y values look like?', 'start': 4881.592, 'duration': 2.061}], 'summary': 'Experiment to determine suitable image types, learning rates, epochs, and their impact on accuracy.', 'duration': 28.566, 'max_score': 4855.087, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/IPBSB1HLNLo/pics/IPBSB1HLNLo4855087.jpg'}], 'start': 4058.548, 'title': 'Deep learning techniques', 'summary': 'Covers deep learning network evolution, setting learning rate, and model training, discussing the evolution of deep learning networks, the importance of learning rate setting, and the process of finding optimal learning rate and training the model, with examples and techniques from recent papers and library implementations.', 'chapters': [{'end': 4316.64, 'start': 4058.548, 'title': 'Deep learning network evolution', 'summary': 'Explains the evolution of a deep learning network through layers, showing how simple features combine to recognize complex patterns, leading to rich capabilities, such as recognizing text, faces, and objects, with just a few layers.', 'duration': 258.092, 'highlights': ["By the time we get to layer five, we've got something that can recognize the eyeballs of insects and birds and something that can recognize unicycle wheels.", "By the third layer, it's actually learnt to recognize the presence of text, petals, and human faces.", 'Layer 2 features include activation by sunset, circular objects, repeating horizontal lines, and corners, showing the combination of layer one features.', 'Layer 1 of the network shows convolutional filters learning diagonal lines, grid patterns, and gradients through gradient descent.']}, {'end': 4602.115, 'start': 4316.64, 'title': 'Setting learning rate in deep learning', 'summary': 'Discusses the importance of setting the learning rate in deep learning, highlighting a new approach to reliably set the learning rate introduced by leslie smith in a 2015 paper, which is built into the fast ai library and involves gradually increasing the learning rate and identifying the point of best improvement.', 'duration': 285.475, 'highlights': ["Leslie Smith's approach to setting the learning rate involves gradually increasing the learning rate and identifying the point of best improvement. Leslie Smith's approach involves gradually increasing the learning rate and identifying the point of best improvement, as demonstrated by plotting the learning rate against the loss to find the point where the graph is dropping the fastest.", 'A new approach to reliably set the learning rate was introduced by Leslie Smith in a 2015 paper and is built into the fast AI library. A new approach to reliably set the learning rate was introduced by Leslie Smith in a 2015 paper and is built into the fast AI library as something called LR finder, which involves gradually increasing the learning rate.', 'The learning rate is a significant factor in gradient descent optimization, affecting the speed and effectiveness of the process. The learning rate significantly impacts the speed and effectiveness of the gradient descent optimization process, with the choice of learning rate being crucial to avoid excessively long or ineffective optimization.']}, {'end': 5193.868, 'start': 4602.115, 'title': 'Learning rate finding and model training', 'summary': 'Discusses the process of finding the optimal learning rate and training the model, emphasizing the use of lrfind method and learning rate scheduler, as well as the importance of determining the number of epochs. it also highlights the need to experiment with different types of images, learning rates, and epochs to understand their impact on model accuracy and overfitting.', 'duration': 591.753, 'highlights': ['The LRFind method is used to find the optimal learning rate for training the model, stopping automatically if the loss worsens, typically stopping before reaching 100%. LRFind method automatically stops training if it notices a significant worsening of the loss, usually before reaching 100%, helping to find the optimal learning rate.', 'The learning rate scheduler object allows plotting the learning rate function to observe its changes over iterations, providing insights into the ideal learning rate. The learning rate scheduler object provides the ability to plot the learning rate function, enabling observation of changes over iterations to determine the ideal learning rate.', 'Experimentation with different learning rate approaches, such as exponential and linear, is conducted to identify the most effective method for model training. The researcher is experimenting with both exponential and linear learning rate approaches to determine the most effective method for model training.', 'The importance of determining the number of epochs for training is emphasized, with the recommendation to run as many epochs as necessary without overfitting or exceeding time constraints. The decision on the number of epochs for training is crucial, with a suggestion to run as many as needed without overfitting or exceeding time constraints.', 'Encouragement to experiment with different types of images, learning rates, and epochs to understand their impact on model accuracy and overfitting, and to gain a deeper understanding of the data and classes. The chapter encourages experimenting with various aspects, including types of images, learning rates, and epochs, to comprehend their impact on model accuracy, overfitting, and data characteristics.']}], 'duration': 1135.32, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/IPBSB1HLNLo/pics/IPBSB1HLNLo4058548.jpg', 'highlights': ['Layer 5 recognizes eyeballs of insects and birds, and unicycle wheels.', 'Layer 3 learns to recognize text, petals, and human faces.', 'Layer 2 features activation by sunset, circular objects, repeating horizontal lines, and corners.', 'Layer 1 shows convolutional filters learning diagonal lines, grid patterns, and gradients.', "Leslie Smith's approach involves gradually increasing the learning rate and identifying the point of best improvement.", 'The learning rate significantly impacts the speed and effectiveness of the gradient descent optimization process.', 'LRFind method automatically stops training if it notices a significant worsening of the loss, usually before reaching 100%.', 'The learning rate scheduler object provides the ability to plot the learning rate function, enabling observation of changes over iterations to determine the ideal learning rate.', 'Experimentation with different learning rate approaches, such as exponential and linear, is conducted to identify the most effective method for model training.', 'The decision on the number of epochs for training is crucial, with a suggestion to run as many as needed without overfitting or exceeding time constraints.', 'The chapter encourages experimenting with various aspects, including types of images, learning rates, and epochs, to comprehend their impact on model accuracy, overfitting, and data characteristics.']}], 'highlights': ['The course consists of seven lessons, most of which will be about a couple of hours long, and is focused on a coding-centric approach to learning deep learning.', 'The model achieved about 99% accuracy in 17 seconds, outperforming the best in the Kaggle competition.', 'The chapter delves into creating Nietzsche philosophy from scratch character by character using a recurrent neural network.', "Demonstration of the model's accuracy with examples of cricket and currency images, achieving nearly perfect results and encouraging experimentation with diverse image sets.", 'The chapter emphasizes the exploration of language and the application of computer vision in the process.', "Deep learning's exponential growth in adoption by major companies like Google and Microsoft, demonstrated through examples such as email replies, real-time language translation in Skype, and medical imaging.", 'Convolutional neural network (CNN) explained using the concept of convolution and a 3x3 kernel for edge detection and image transformation.', 'Layer 5 recognizes eyeballs of insects and birds, and unicycle wheels.', "Leslie Smith's approach involves gradually increasing the learning rate and identifying the point of best improvement.", 'The learning rate significantly impacts the speed and effectiveness of the gradient descent optimization process.']}