4 ms·
Goodfellow's [1] comes closest I'd say, but since deep learning is still moving very fast and lacks a solid formal grounding (unlike most other machine learning
by blackbear_ 6y ago
Goodfellow's [1] comes closest I'd say, but since deep learning is still moving very fast and lacks a solid formal grounding (unlike most other machine learning methods) it's not so easy to find a comprehensive book yet.
[1] https://www.deeplearningbook.org/ https://www.deeplearningbook.org/
- mrfusion 6y agoWill that teach me transformers too?
- nmfisher 6y agoNo (at least, not the version I read a few years ago). But transformers are a specific neural network architecture, so I’d still recommend the book for the fundamentals around backpropagation, activation & loss functions, etc. Once you’re comfortable with neural networks (and the notation), the “Attention is all you need” paper is fairly accessible.
- mrfusion 6y ago> Once you’re comfortable with neural networks (and the notation) I think I am actually. And I read attention is all you need. (Well half of it) and it didn’t seem to delve into how they work.
- blackbear_ 6y agoIt really does, read section 3 carefully. As for why they work so well, that's still an open question.
- AlexCoventry 6y agoFor transformers, try this (also links to a video introduction): http://jalammar.github.io/illustrated-transformer/ http://jalammar.github.io/illustrated-transformer/
- huitzitziltzin 6y agoCovered in "Dive Into Deep Learning," which another commenter has recommended. https://d2l.ai/chapter_attention-mechanisms/transformer.html https://d2l.ai/chapter_attention-mechanisms/transformer.html