Xiusi Chen

Perception-Aware Policy Optimization for Multimodal Reasoning

How Far Can Unsupervised RLVR Scale LLM Training?

RM-R1: Reward Modeling as Reasoning

Beyond Log Likelihood: Probability-Based Objectives for Supervised Fine-Tuning across the Model Capability Continuum

Teaching Language Model to Act Efficiently

ToolRL: Reward is All Tool Learning Needs

Large Scale Knowledge Washing

Self-Updatable Large Language Models by Integrating Context into Model Parameters

Strategist: Self-improvement of LLM Decision Making via Bi-Level Tree Search