8 papers
Understanding and Preventing Entropy Collapse in RLVR with On-Policy Entropy Flow Optimization
Huimin Xu, Shuai Zhao, Xiaobao Wu +1
Reinforcement learning with verifiable rewards (RLVR) has become an effective paradigm for improving the reasoning ability of large language models. However, widely used RLVR algor…
EduStory: A Unified Framework for Pedagogically-Consistent Multi-Shot STEM Instructional Video Generation
Xinyi Wu, Jayant Teotia, Shuai Zhao +1
Long-horizon video generation has advanced in visual quality, yet existing methods still struggle to maintain knowledge consistency and coherent pedagogical narratives across multi…
From Stimuli to Minds: Enhancing Psychological Reasoning in LLMs via Bilateral Reinforcement Learning
Yichao Feng, Haoran Luo, Lang Feng +2
Large Language Models show promise in emotion understanding, social reasoning, and empathy, yet they struggle with psychologically grounded tasks that require inferring implicit me…
CutPaste&Find: Efficient Multimodal Hallucination Detector with Visual-aid Knowledge Base
Cong-Duy Nguyen, Xiaobao Wu, Duc Anh Vu +3
Large Vision-Language Models (LVLMs) have demonstrated impressive multimodal reasoning capabilities, but they remain susceptible to hallucination, particularly object hallucination…
AntiLeakBench: Preventing Data Contamination by Automatically Constructing Benchmarks with Updated Real-World Knowledge
Xiaobao Wu, Liangming Pan, Yuxi Xie +7
Data contamination hinders fair LLM evaluation by introducing test data into newer models' training sets. Existing studies solve this challenge by updating benchmarks with newly co…
Unlearning Backdoor Attacks for LLMs with Weak-to-Strong Knowledge Distillation
Shuai Zhao, Xiaobao Wu, Cong-Duy Nguyen +4
Parameter-efficient fine-tuning (PEFT) can bridge the gap between large language models (LLMs) and downstream tasks. However, PEFT has been proven vulnerable to malicious attacks.…