Background: I know how to program (Python, C, Racket). I have done small projects and am confident on that front that if I understand the inner workings of something, then I can build that thing using code.<p>Now, I want to learn Deep Learning with emphasis on computer vision although it wouldn't hurt to learn other applications. But my focus is on core deep learning and computer vision.<p>I want to understand research papers, implement them from scratch (using something like PyTorch) and build new models to improve upon SOTA performance. By 'master' I mean all of the above.<p>How to go about that? Which courses, books, tutorials are the best to focus on? I know a bit of linear algebra and calculus and hope to pick up things as I go on.
This is the best starting point: <a href="http://neuralnetworksanddeeplearning.com/" rel="nofollow">http://neuralnetworksanddeeplearning.com/</a><p>Then:<p><a href="https://m.youtube.com/playlist?list=PL5-TkQAfAZFbzxjBHtzdVCWE0Zbhomg7r" rel="nofollow">https://m.youtube.com/playlist?list=PL5-TkQAfAZFbzxjBHtzdVCW...</a><p>and/or:<p><a href="https://www.fast.ai/posts/2022-07-21-dl-coders-22.html" rel="nofollow">https://www.fast.ai/posts/2022-07-21-dl-coders-22.html</a><p>p.s. I could only dream about such learning resources when I started as a DL researcher in 2013.
I would start here: <a href="https://www.deeplearningbook.org/" rel="nofollow">https://www.deeplearningbook.org/</a><p>If you already know all of the math in Part I, great, but if you don't you need to study it carefully if you want to understand what is going on. If you just want to code, there are shorter paths, but if you really want to understand the theory, you need to master the Linear Algebra and Probability material in the first part of that book. Parts II and III give a solid foundation in DL itself.
Yi Ma’s book and slides seems ok to me. In sense that ML is like Signal Processing and Control Theory. Though I think he is somewhat against “the bitter lesson”<p><a href="https://book-wright-ma.github.io/" rel="nofollow">https://book-wright-ma.github.io/</a>
<a href="http://www.incompleteideas.net/IncIdeas/BitterLesson.html" rel="nofollow">http://www.incompleteideas.net/IncIdeas/BitterLesson.html</a>
Implement and train AlexNet in C without using any external libraries. Once you've done that, do the same for a text-generation LSTM. Once you've done that, write a GAN. And so on.