Elvis Dohmatob

Associate Professor@Concordia University·加拿大·OpenReview

研究方向

neural scaling laws · synthetic data · random matrix theory · adversarial examples · deep learning theory · determinantal point processes · robust optimization · learning theory · convex optimization

7.5

Understanding Softmax Attention Layers:\\ Exact Mean-Field Analysis on a Toy Problem

NeurIPS 2025Poster

一作

6.4

Improving the Scaling Laws of Synthetic Data with Deliberate Practice

ICML 2025Oral

三作

5.8

The Pitfalls of Memorization: When Memorization Hurts Generalization

合作者 (20)

Reyhane Askari-Hemmat

2 篇

Arjun Subramonian

2 篇

Adriana Romero-Soriano

Elvis Dohmatob

Why Less is More (Sometimes): A Theory of Data Curation

Egalitarian Gradient Descent: A Simple Approach to Accelerated Grokking

Strong Model Collapse

An Effective Theory of Bias Amplification

Beyond Model Collapse: Scaling Up with Synthesized Data Requires Verification

Understanding Softmax Attention Layers:\\ Exact Mean-Field Analysis on a Toy Problem

Improving the Scaling Laws of Synthetic Data with Deliberate Practice

The Pitfalls of Memorization: When Memorization Hurts Generalization