Data-driven models learned often struggle to generalize due to widespread subpopulation shifts, especially the presence of both spurious correlations and group imbalance (SC-GI).

We conduct a novel theoretical analysis to demonstrate CODA's

Thispaperproposes a new quasi-attention that is compositional in nature, i.e., learning whether to add, subtract or nullify a certain vector when learning representations. This is strongly contrasted with vanilla attention, which simply re-weights input tokens.

Neural Information Processing Systems http://nips.cc/

Data augmentation plays a key role in modern machine learning pipelines.

To meaningfully regularize the model's decision boundary, CODA introduces virtual unknown classes