Optimization for ML at Rice University (CS) led by Associate Prof. Anastasios Kyrillidis - Efficient training methods, non-convex optimization, and more.
Houston, Texas
Joined December 2020
- Tweets509
- Following236
- Followers1.7K
- Likes911
Long time, no post (everyone is "running" freaking crazy these days...)
Personal thoughts from chats with colleagues at Rice: Will AI hurt mathematics more than computer science?
My answer (probably I'm wrong): AI will change how we do math. It will not end it. 🧵
(For Rice colleagues: a coffee conversation with no single right answer. Excellent excuse to have a cup of coffee ☕)
@TaoistTerence uses AI tools openly in his work. I read that as an invitation to all of us.
Nobody has the map yet, not for research and not for teaching.
It has never been easier to skip the meeting: an LLM iterates faster than four people in a room.
But the whiteboard leap (the wrong idea that becomes the right one at minute 20) never came out of a chat window.
My recipe: AI to prepare, the room to decide. What's yours?
Same matrix. Same condition number. Three iterations, or three thousand.
Every quantum linear solver for Ax = b quotes its cost in κ. New post on what κ is actually measuring
and why "easy instance" means two different things to a quantum solver and to conjugate
gradients. 🧵
Random thoughts that everyone has during peer-reviewed rebuttal period in a top AI conference:
- Reviewers use AI to get a first idea about the paper. AI inserts their own bias about the paper;
- Reviewers use AI to write the review but do not step their foot on what AI proposes
- Authors read and throw reviews into AI and get answers how to tackle them
- Authors write rebuttals based on AI's response and with the help of AI
- Reviewers see the rebuttal and throw it in AI to get answers. Initial bias (positive or negative) is kept: you can always find reasons to accept or reject a paper; definitely, for sure, about rejecting a paper.
What does this leave me (forget about "peer review is rigged or destroyed") as a taste? The only papers that are reasonable to accept without a doubt are those that are based on mathematical foundations ("here is an idea, here is the proof, the proof is correct, the experiments support claims, we should care whether the method is better only if there are strict fair rules of the comparison"). Someone could say "Yes, but these can also be written by AI". Agreed, but again that's the only thing we are left; anything else is pure lack about what is the criterion to get in.
Anything else ("here is a small change of an algorithm, here is tweak I made, look what happens if i change the order of the data,..." etc) is amenable to a matter of taste from the reviewer; and now this taste is passed into what AI models think.
What I tell my students: not to be dissappointed, writing papers that they like is the essence, not publishing. This is where we got (and AI was not the only reason; it is the greediness of people writing papers like there is no tomorrow, like having 10+ accepted papers in conferences (which means that they submitted 30-40)).
Thinking about new ways to disseminate work is imperative.
Disclaimer: I'm not less guilty of any of the above.
This is a joint effort: Tirthak Patel, @NaiHuiChia2, Shengxi Huang, @QuantumHazzard, Leonardo Dueñas-Osorio, Guido Pagano, Brandon Ruzic, Clena Abuan, @QuantinuumQC, @IQMquantum
@ENERGY
news.rice.edu/news/2026/rice…
