7 papers
Decoupling Knowledge Forgetting and Refusal Learning in MLLM Unlearning
Haodong Chen, Shengtao Wen, Dong Liang +2
Current evaluations of multimodal unlearning typically focus on whether a model stops outputting target information. This surface-level criterion, however, masks two critical failu…
Before the Token Commits: Trajectory-Level Benchmarking of Visual Hallucinations in Diffusion VLMs
Yadong Wang, Siping Yue, Yu Tian +2
Multimodal diffusion language models generate responses by iteratively unmasking tokens, making each answer the endpoint of a multi-step trajectory rather than an immediate commitm…
Knowing but Not Saying: Preventing Factual Access Failures in LLM SFT via Recall-Anchored Distillation
Haodong Chen, Yadong Wang, Shengtao Wen +2
Supervised fine-tuning (SFT) can degrade factual behavior outside the target domain. This degradation is often described as catastrophic forgetting, yet open-ended factual failures…
Compress the Easy, Explore the Hard: Difficulty-Aware Entropy Regularization for Efficient LLM Reasoning
Qin-Wen Luo, Sheng Ren, Xiang Chen +4
Chain-of-Thought (CoT) has substantially empowered Large Language Models (LLMs) to tackle complex reasoning tasks, yet the verbose nature of explicit reasoning steps incurs prohibi…
Beyond Dense States: Sparse Transcoders as Causally Testable Operators for LLM Latent Reasoning
Yadong Wang, Haodong Chen, Yu Tian +3
Latent reasoning reduces the token-generation cost of chain-of-thought reasoning by replacing explicit intermediate tokens with continuous latent transitions. However, existing lat…
MultiMedEdit: A Scenario-Aware Benchmark for Evaluating Knowledge Editing in Medical VQA
Shengtao Wen, Haodong Chen, Yadong Wang +6
Knowledge editing (KE) provides a scalable approach for updating factual knowledge in large language models without full retraining. While previous studies have demonstrated effect…
Reflect then Learn: Active Prompting for Information Extraction Guided by Introspective Confusion
Dong Zhao, Yadong Wang, Xiang Chen +6
Large Language Models (LLMs) show remarkable potential for few-shot information extraction (IE), yet their performance is highly sensitive to the choice of in-context examples. Con…