Noam Itzhak Levi

Postdoc@EPFL - EPF Lausanne·瑞士·OpenReview

研究方向

RMT · emergent phenomena · deep learning · scaling laws · grokking · Theoretical Physics · Deep Learning · Theoretical Particle Physics · Early Universe Cosmology

6.0

Pretraining Scaling Laws for Generative Evaluations of Language Models

ICLR 2026Poster

二作

5.3

Learning Shrinks the Hard Tail: Training‑Dependent Inference Scaling in a Solvable Linear Model

合作者 (16)

Noam Itzhak Levi

Pretraining Scaling Laws for Generative Evaluations of Language Models

Learning Shrinks the Hard Tail: Training‑Dependent Inference Scaling in a Solvable Linear Model

Evaluating the Robustness of Chinchilla Compute-Optimal Scaling

Ascent Fails to Forget

Probing the Latent Hierarchical Structure of Data via Diffusion Models

Grokking at the Edge of Linear Separability

A Simple Model of Inference Scaling Laws

The Underlying Universal Statistical Structure of Natural Datasets

Grokking at the Edge of Linear Separability

A Phase Transition Induces Catastrophic Overfitting in Adversarial Training