12 papers
Measure, Don't Optimize: Forecasting Recovery in LLM Unlearning
Zirui Song, Huaxing Liu, Xiang Wang +8
Prior white-box studies show that large language models can retain latent traces of target knowledge after unlearning, even when the knowledge is no longer expressed in their outpu…
One Ranking, Any Budget: Matryoshka Evidence-to-Context Frame Selection for Long-Video Understanding
Wang Chen, Yu Chen, Xiang Wang +3
Frame selection is essential for applying Large Multimodal Models (LMMs) to long videos due to severe frame redundancy and limited context windows. Since the appropriate frame budg…
CAVE: Competence-Aware Visual Boundary Evidence Alignment for Video Temporal Grounding
Wei Jia, Zhicong Lu, Yu Chen +6
Large vision-language models (LVLMs) have achieved substantial performance gains in Video Temporal Grounding (VTG) through reinforcement learning (RL). However, existing methods pr…
Credit the Right Box: Marginal Contribution Assignment for Structured Visual Perception
Xinheng Han, Jianfei Wang, Yu Chen +4
Multimodal Large Language Models (MLLMs) are increasingly expected to solve structured perception tasks that require visual recognition, language-to-object binding, object cardinal…
Inside the Unfair Judge: A Mechanistic Interpretability Account of LLM-as-Judge Bias
Zixiang Xu, Sixian Li, Huaxing Liu +4
The paper investigates how biases in large language models used as judges are reflected in their hidden activations, identifying low-dimensional subspaces that encode bias and show…
Do Multimodal Agents Really Benefit from Tool Use? A Systematic Study of Capability Gains
Garvin Guo, Donglei Yu, Yu Chen +6
Tool-augmented multimodal agents show strong benchmark gains, often taken as evidence that agents have learned to use tools. We argue that this interpretation can be premature: a t…