Comment by PAClearner | Hacker News Reader

Comment by PAClearner | Hacker News Reader

Parent Full thread

Well I would be happy to provide some context. I just finished my first year of CS Phd in ML (more on the theory side) and I really like it. I think most of the places you would want to do a post doc in CS are probably going to be moderately high CoL. My phd is in a place with pretty low CoL (but a still a top 10-top 20 school (depending on who you ask) ) so the graduate stipend goes reasonably far.

The other thing to note in ML is that it seems like a few people go to industry research labs for a few years i.e MSR/FAiR/google brain and then come back to the academy since there are industry roles that involve research and publication. for instance moritz hardt.

my personal plan for the first 3 years of grad school is to work really hard and try to keep both academia and industry open and after year 3 evaluate the number of publications I have and my current skill set to see if I can make it in the academy or shift more towards industry.

I think the biggest factor I would comment on is look very closely about what jobs the graduated students from the department you matriculate at AND more importantly the professor you want to work with go on to do post Phd. There are a lot of naysayers in this thread about the risks of an academic career and I share those concerns but I felt a lot more comfortable taking the plunge after I looked at the career record of the graduated students of my advisor. They were all either tenure track or had good industry positions.

edit: if your advisor has collaborators in industry groups I think it is pretty straight forward to get an industry gig.

When evaluating the warnings from naysayers you have to keep in mind that CS is quite an outlier as far as backup career prospects go. I made it all the way to CS postdoc and every step of the way I had to keep swatting away industry recruiters waving wads of cash at me. I finally made the leap for other reasons but it was effortless. I think this is absolutely not how it works in other disciplines.

One exception I can think of is what I call "closet programmers," which are folks that work in various areas which rely on software such as experimental physics, astronomy, molecular biology. and end up mostly doing programming because they love it. We have a bunch of engineers like that and they are all excellent :-)

100% correct. I think we are both very lucky in that we are able to do fun science and chase our intellectual interests with a realistic and still fun safety net.

Also those closet programmers are always really fun to talk to since their problems and culture are breath of fresh air.

Thanks for the context. That sounds like a good plan to me regarding post-doc locations. I am also interested in theory side of ML. What areas of mathematics should one learn really well that apply to the theory side? What blogs, papers, books would you point one to to learn the theory side more? To your knowledge are their applications of abstract algebra to ML? If so, what areas of algebra apply & what problems do they solve?

I could rant about this for hours. I actually just went to a defense for a deep learning paper that had a ton of abstract algebra. I am honestly not really a fan of deep learning and algebra because all the papers to me like- seem to stop at describing some really basic feedforward network as some really specific mathematical structure but these theories a) provide very little explanation of empirical phenomena b) provide no new directions of research in terms of like useful network architectures.

I haven't really come across algebra in machine learning other than people applying it to deep learning.

i.e. https://arxiv.org/pdf/1802.03690.pdf

ie. https://icml.cc/Conferences/2018/Schedule?showEvent=2048

I don't personally find papers like this valuable but idk I have never really enjoyed abstract algebra.

For areas of mathematics to do theory in ML (and to do ML more generally!)

-probability/concentration/hoeffding bounds [the PAC model] [Key]

-linear algebra [key]

-optimization [key]

for books

-understanding machine learning by shai ben david

This book is nice since it really balances theory with a more practical understanding.

-An Introduction to Computational Learning Theory by kearns is a classic [low priority]. this is fun since the proofs are simple and deep but is very very far away from practical algorithms.

-convex optimization by boyd

Course Notes:

[I think a good alternative to blogs is stalking course notes for other schools-they are very often public.]

- http://ttic.uchicago.edu/~avrim/MLT18/index.html

good learning theory course by avrim blum who is a big deal in learning theory and theory.

- tim roughgardens notes are a blessing for algorithms and theory [seriously he should have a patreon or something]

https://theory.stanford.edu/~tim/notes.html

Blogs:

-http://www.argmin.net/

this is ben recht's blog and is filled with ML wisdom.

-https://blogs.princeton.edu/imabandit/ not quite learning theory but a lot of ML adjacent stuff

I don't read many blogs as I should tbh so other people can give better advice

VIDEOS https://www.youtube.com/channel/UCW1C2xOfXsIzPgjXyuhkw9g

This is the simons institute youtube channel. probably the best single location for recordings of TALKS in computer science-good amount of ML talks.

https://simons.berkeley.edu/videos

I don't see a way to respond to your latest reply, but thank you for the recommendations! I'll take a look at them.

Reply on news.ycombinator.com