Lecturer, Huazhong University of Science and Technology
2 papers at NeurIPS 2025
To address inefficiency from excessive visual tokens in LVLMs, we propose an information-flow perspective revealing dynamic redundancy emergence and introduce a method aligned with the model’s inherent behavior, outperforming all existing approaches.
This paper introduces an IRM-free, norm-guided invariant distribution criterion for causal subgraph discovery in graph OOD generalization.
