1

We systematically evaluate and improve negation understanding in VLMs, first exposing severe affirmation bias and then proposing a data-centric fix.

We introduce FedMedICL, a unified framework and benchmark to evaluate the impact of simultaneous distribution shifts (label, demographic, and temporal) on federated learning for medical imaging.

Current evaluations of Continual Learning (CL) methods typically assume that there is no constraint on training time and computation. …

We leverage learnable tokens and large-scale pretrained models to mitigate forgetting in video class incremental learning.

Continual learning (CL) is under-explored in the video domain. The few existing works contain splits with imbalanced class …