5 papers across 3 sessions
We leverage the closed-form formulation of flow matching to understand its generalization
We propose CLIPTTA, a contrastive test-time adaptation method for CLIP that improves both accuracy and OOD detection in closed- and open-set settings.
This paper provides the first Wasserstein convergence guarantees for critically damped Langevin diffusions.
We analyze imbalanced training loss, showing that gradient descent dynamics can gradually reduce bias and recover minority-specific features with longer training.