OTTER: Improving Zero-Shot Classification via Optimal Transport

doi:doi:10.57702/85fh6h5w

OTTER: Improving Zero-Shot Classification via Optimal Transport

Zero-shot models suffer due to artifacts inherited from pretraining. A particularly detrimental artifact, caused by unbalanced web-scale pretraining data, is mismatched label distribution. Existing approaches that seek to repair the label distribution are not suitable in zero-shot settings, as they have incompatible requirements such as access to labeled downstream task data or knowledge of the true label balance in the pretraining distribution.

BibTex:

@dataset{Changho_Shin_and_Jitian_Zhao_and_Sonia_Cromp_and_Harit_Vishwakarma_and_Frederic_Sala_2025,
    abstract = {Zero-shot models suffer due to artifacts inherited from pretraining. A particularly detrimental artifact, caused by unbalanced web-scale pretraining data, is mismatched label distribution. Existing approaches that seek to repair the label distribution are not suitable in zero-shot settings, as they have incompatible requirements such as access to labeled downstream task data or knowledge of the true label balance in the pretraining distribution.},
    author = {Changho Shin and Jitian Zhao and Sonia Cromp and Harit Vishwakarma and Frederic Sala},
    doi = {10.57702/85fh6h5w},
    institution = {No Organization},
    keyword = {'Image Classification', 'Label Distribution', 'Optimal Transport', 'Text Classification', 'Zero-Shot Learning'},
    month = {jan},
    publisher = {TIB},
    title = {OTTER: Improving Zero-Shot Classification via Optimal Transport},
    url = {https://service.tib.eu/ldmservice/dataset/otter--improving-zero-shot-classification-via-optimal-transport},
    year = {2025}
}