Full Professor, Nanjing University of Aeronautics and Astronautics
1 paper at NeurIPS 2025
We introduce a representation-level counterfactual framework to mitigate co-occurrence bias in vision–language models without retraining or prompt engineering.
