Tengyu Ma

Stanford University@Stanford University·OpenReview

研究方向

theory · optimization · machine learning theory · non-convex optimization · generalization theory · deep learning theory · reinforcement learning theory

5.5

Fantastic Pretraining Optimizers and Where to Find Them

ICLR 2026Poster

三作

7.2

STP: Self-play LLM Theorem Provers with Iterative Conjecturing and Proving

ICML 2025Poster

二作

6.0

Understanding Warmup-Stable-Decay Learning Rates: A River Valley Loss Landscape View

合作者 (20)

David Leo Wright Hall

Tengyu Ma

Fantastic Pretraining Optimizers and Where to Find Them

STP: Self-play LLM Theorem Provers with Iterative Conjecturing and Proving

Understanding Warmup-Stable-Decay Learning Rates: A River Valley Loss Landscape View

Larger language models do in-context learning differently

Non-Asymptotic Length Generalization

Formal Theorem Proving by Rewarding LLMs to Decompose Proofs Hierarchically