Hacker News Reader

Comment by noelwelsh | Hacker News Reader

I'm confused as to the intended audience for this. It's not a textbook. In the bits I looked at it doesn't attempt to teach the material, just banging out definitions and theorems.

Also, the table of contents doesn't have hyperlinks to the actual content, which makes navigation a massive annoyance. Similarly, there aren't hyperlinks to the references.

An impressive amount of work went into this, but why?

View on HN

The book is still in progress [0]. For example, the introduction has yet to be written, which presumably would answer some of these questions about the intended audience.

The author also has books [1] on tangentially related subjects, several with a good amount of overlap in material.

[0] https://www.cis.upenn.edu/~jean/gbooks/geomath.html

[1] https://www.cis.upenn.edu/~jean/gbooks/home.html

> just banging out definitions and theorems

This is pretty standard for graduate-level math texts, though. That might say something about their intended audience.

I feel this is good foundation in CS and interested into more advanced ML and math. At least for me I feel excited when reading the toc since I have touched many of those concept over the years but never went deeper

The space of permissible transformations is of great interest since moving beyond linear combination alla ANNs has lead to great performance strides in deep learning?

It's just a reference, I wouldn't view it as teaching material.

Its quite useful to look things up in.

Yep, this is an extended survey of interrelated concepts organized according to increasing complexity.

Similar to textbooks in the organizational principle, but differentiated in substance in that it assumes familiarity with the underlying topics and focuses on the relationships between the concepts rather than the concepts themselves. Good for practitioners I think.

It is a textbook.

Reply on news.ycombinator.com