Juno KIM

@junokim_ai

PhD student in EECS @UCBerkeley working on deep learning theory; previously SNU Math, UTokyo IST; @GoogleDeepMind intern

junokim1.github.io

Joined January 2024

140 Following

667 Followers

Replies Reposts Media

Pinned
Juno KIM
@junokim_ai
Oct 1
What's the best way to solve linear regression? (No, it’s not least squares.) In our new paper, we show that principal component regression (PCR) beats, up to constants, every monotone spectral filter (incl. gradient descent & ridge regression) on every problem instance! (1/8)
12
Juno KIM
@junokim_ai
Oct 27
Very timely project! A lot of the mathematics questions were technically well posed but somewhat derivative or "second order" (but then again, so is a lot of human research). It would be cool to push this to ML theory problems with experimental components.
Fan Nie
@FanNie1208
Oct 27
AI is now taking on some of the hardest open problems that scientists have posed. But research is broader than solving problems. Can AI also propose research questions that experts find genuinely interesting? We built After42 to find out. Experts judged 45 of its 135 final
Juno KIM
@junokim_ai
Sep 27
Accepted to #neurips2026 following an oral presentation in the ICML high-dimensional learning dynamics workshop!
Juno KIM
@junokim_ai
Mar 30
Excited to share our new paper on sharp capacity scaling of the Muon optimizer! Joint work with @EshaanNichani Denny Wu @albertobietti @jasondeanlee: arxiv.org/abs/2603.26554 (1/7)
1
Juno KIM
@junokim_ai
Aug 23
The maximum capacity n of a d x d linear associative memory (LAM) for perfect retrieval is given by d^2=2n*log n. Joint work with Nicholas Barnfield, @EshaanNichani, @jasondeanlee, Yue M. Lu: arxiv.org/abs/2605.05189 (1/4)
arxiv.org
Sharp Capacity Thresholds in Linear Associative Memory: From Top-1...
How many key-value associations can a $d\times d$ linear memory store? The answer depends not only on the $d^2$ degrees of freedom in the memory matrix, but also on the retrieval criterion. Under...
4
Juno KIM
@junokim_ai
Jul 7
Our new paper at #ICML2026 studies why simple on-policy preference learning (e.g., online DPO) can outperform their offline counterparts! Joint work with Jihun Yun @jasondeanlee @kwangsungjun Presenting tomorrow (July 8) 5pm, Hall A #4514! arxiv.org/abs/2601.08421 1/5
1

Juno KIM

Juno KIM’s posts