Fuxiang Zhang

Skywork-Reward-V2: Scaling Preference Data Curation via Human-AI Synergy

Uncertainty-Aware Tree Search for Efficient LLM Reasoning

REAR: Scalable Test-time Preference Realignment through Reward Decomposition

Adversarial Test Case Generation via Reinforcement Learning Extends Scaling Laws

Incentivizing LLMs to Self-Verify Their Answers

Q-Adapter: Customizing Pre-trained LLMs to New Preferences with Forgetting Mitigation

Multi-Agent Imitation by Learning and Sampling from Factorized Soft Q-Function