OTTER: Effortless Label Distribution Adaptation of Zero-shot Models
This addresses label distribution adaptation for zero-shot models, enabling better performance without needing labeled downstream data, though it is incremental as it builds on existing optimal transport methods.
The paper tackles the problem of mismatched label distribution in zero-shot models caused by unbalanced pretraining data, introducing a lightweight optimal transport approach that improves accuracy by 4.8% and 15.9% on average in image and text classification tasks.
Popular zero-shot models suffer due to artifacts inherited from pretraining. One particularly detrimental issue, 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 mismatching requirements, such as needing access to labeled downstream task data or knowledge of the true label balance in the pretraining distribution. We sidestep these challenges and introduce a simple and lightweight approach to adjust pretrained model predictions via optimal transport. Our technique requires only an estimate of the label distribution of a downstream task. Theoretically, we characterize the improvement produced by our procedure under certain mild conditions and provide bounds on the error caused by misspecification. Empirically, we validate our method in a wide array of zero-shot image and text classification tasks, improving accuracy by 4.8% and 15.9% on average, and beating baselines like prior matching -- often by significant margins -- in 17 out of 21 datasets.