Class Distribution Shifts in Zero-Shot Learning: Learning Robust Representations
Recommended citation: Slavutsky, Y., & Benjamini, Y. (2024). "Class Distribution Shifts in Zero-Shot Learning: Learning Robust Representations." Conference on Neural Information Processing Systems (NeuRIPS).
Class distribution shifts challenge zero-shot classifiers, which rely on representations from training classes but are tested on unseen ones. We propose a model that assumes an unknown class attribute causes the shift and present an algorithm to learn robust representations where ERM would fail.