Zhiyuan Li

On Powerful Ways to Generate: Autoregression, Diffusion, and Beyond

A Tale of Two Geometries: Adaptive Optimizers and Non-Euclidean Descent

The Information Bottleneck of Chain-of-Thought and How Latent CoT Overcomes It

Provable Benefits of Sinusoidal Activation for Modular Addition

Adam Exploits $\ell_\infty$-geometry of Loss Landscape via Coordinate-wise Adaptivity

PENCIL: Long Thoughts with Short Memory

Chain-of-Thought Provably Enables Learning the (Otherwise) Unlearnable

Reasoning with Latent Thoughts: On the Power of Looped Transformers

On Learning Verifiers and Implications to Chain-of-Thought Reasoning

A Coefficient Makes SVRG Effective

Weak-to-Strong Generalization Even in Random Feature Networks, Provably

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

Non-Asymptotic Length Generalization

Shift is Good: Mismatched Data Mixing Improves Test Performance

Structured Preconditioners in Adaptive Optimization: A Unified Analysis