Simon Shaolei Du

Improving Human-AI Coordination through Online Adversarial Training and Generative Models

Convergence Dynamics of Over-Parameterized Score Matching for a Single Gaussian

Global Convergence of Four-Layer Matrix Factorization under Random Initialization

Chasing Moving Targets with Online Self-Play Reinforcement Learning for Safer Language Models

Unregularized Linear Convergence in Zero-Sum Game from Preference Feedback

PrefDisco: Benchmarking Proactive Personalized Reasoning

Policy-Based Trajectory Clustering in Offline Reinforcement Learning

Understanding the Performance Gap in Preference Learning: A Dichotomy of RLHF and DPO

Spurious Rewards: Rethinking Training Signals in RLVR

Cross-environment Cooperation Enables Zero-shot Multi-agent Coordination

Sharp Gap-Dependent Variance-Aware Regret Bounds for Tabular MDPs

Reinforcement Learning for Reasoning in Large Language Models with One Training Example

A Minimalist Example of Edge-of-Stability and Progressive Sharpening

Extragradient Preference Optimization (EGPO): Beyond Last-Iterate Convergence for Nash Learning from Human Feedback

Deployment Efficient Reward-Free Exploration with Linear Function Approximation

Transformers are Efficient Compilers, Provably

LoRe: Personalizing LLMs via Low-Rank Reward Modeling

Understanding the Gain from Data Filtering in Multimodal Contrastive Learning

Highlighting What Matters: Promptable Embeddings for Attribute-Focused Image Retrieval

Minimax Optimal Regret Bound for Reinforcement Learning with Trajectory Feedback

The Crucial Role of Samplers in Online Direct Preference Optimization

Minimax Optimal Regret Bound for Reinforcement Learning with Trajectory Feedback

Multi-Agent Reinforcement Learning from Human Feedback: Data Coverage and Algorithmic Techniques

SHARP: Accelerating Language Model Inference by SHaring Adjacent layers with Recovery Parameters

Deployment Efficient Reward-Free Exploration with Linear Function Approximation

On Erroneous Agreements of CLIP Image Embeddings

Transformers are Efficient Compilers, Provably