5 papers
Calibration-Aware Policy Optimization for Reasoning LLMs
Ziqi Wang, Xingzhou Lou, Meiqi Wu +2
Group Relative Policy Optimization (GRPO) enhances LLM reasoning but often induces overconfidence, where incorrect responses yield lower perplexity than correct ones, degrading rel…
Unveiling Chain of Step Reasoning for Vision-Language Models with Fine-grained Rewards
Honghao Chen, Xingzhou Lou, Xiaokun Feng +2
Chain of thought reasoning has demonstrated remarkable success in large language models, yet its adaptation to vision-language reasoning remains an open challenge with unclear best…
Uncertainty-aware Reward Model: Teaching Reward Models to Know What is Unknown
Xingzhou Lou, Dong Yan, Wei Shen +3
Reward models (RMs) are essential for aligning large language models (LLM) with human expectations. However, existing RMs struggle to capture the stochastic and uncertain nature of…
Reward-Robust RLHF in LLMs
Yuzi Yan, Xingzhou Lou, Jialian Li +6
As Large Language Models (LLMs) continue to progress toward more advanced forms of intelligence, Reinforcement Learning from Human Feedback (RLHF) is increasingly seen as a key pat…
SPO: Multi-Dimensional Preference Sequential Alignment With Implicit Reward Modeling
Xingzhou Lou, Junge Zhang, Jian Xie +3
Human preference alignment is critical in building powerful and reliable large language models (LLMs). However, current methods either ignore the multi-dimensionality of human pref…