Assistant Professor, Mohamed bin Zayed University of Artificial Intelligence
2 papers at NeurIPS 2025
We show that visual input induces an activation shift that weakens VLM safety and propose a calibration method to mitigate this effect.
We introduce CLIMB, a comprehensive benchmark and empirical study of 29 class-imbalanced learning methods on 73 real-world tabular datasets, revealing key insights into method performance, efficiency, and robustness.
