News for hackers. Stuff that matters.
Re:Mathematical Introduction to Deep Learning: Methods, Impl... by two_in_one on Monday January 01, 2024 @10:25PM (#38837742)
It's hard to call comprehensive. Transformers - one page. A picture would be nice. No "prompt engineering", no "double deep". In fact words "prompt" and "double" aren't used at all. "Recognition" is used only once outside of bibliography just for reference. Looks like theory will not catch up with practice any time soon. With looming singularity it's bit worrying.
Re:Mathematical Introduction to Deep Learning: Methods, Impl... by fourier456 on Monday January 01, 2024 @10:34PM (#38837800)
This just isn’t really a good book. One sign; it’s an arXiv book.
Re:Mathematical Introduction to Deep Learning: Methods, Impl... by tnecniv on Tuesday January 02, 2024 @01:24AM (#38838765)
Right in the title it explains it’s a book on theory. “Prompt engineering” doesn’t really fit in any theoretical framework I’m aware of and, while I also like graphics, most theory publications are light on them. You might be looking for a different kind of book, which is fine, but I think the content matches the title.

Also, a book on theory is going to lag quite a bit in terms of topics. The general process is that people discover something new and interesting empirically and publish articles on it. Other people develop theory explaining why that things work and publish articles on it. Once the theory gets crystallized, big ideas get distilled into a book.

Re:Mathematical Introduction to Deep Learning: Methods, Impl... by nephanth on Tuesday January 02, 2024 @11:04AM (#38843038)
It's an introductory book though? I don't think it aims at being comprehensive

That said, i do agree that more on transformers would be nice since they're becoming quite central in every field of machine learning.

Prompt engineering is extremely new, vastly empirical, and theory on is still only beginning (though i do remember seeing some nice papers passing). It would probably be a mistake to include it in an introductory book

I have never heard of "double deep", what is that?