Generalized Logit Adjustment: Calibrating Fine-tuned Models by Removing Label Bias in Foundation Models
This addresses a fundamental flaw in pre-training for researchers and practitioners using foundation models, offering a method to improve performance across diverse tasks, though it is incremental as it builds on existing debiasing techniques.
The paper tackles the problem of inherent biases in foundation models like CLIP, which arise from imbalanced training data and affect fine-tuning and ensembling, by proposing the Generalized Logit Adjustment (GLA) method to remove label bias, resulting in accuracy gains such as 1.5 percentage points on ImageNet and up to 4.6 percentage points on few-shot datasets.
Foundation models like CLIP allow zero-shot transfer on various tasks without additional training data. Yet, the zero-shot performance is less competitive than a fully supervised one. Thus, to enhance the performance, fine-tuning and ensembling are also commonly adopted to better fit the downstream tasks. However, we argue that such prior work has overlooked the inherent biases in foundation models. Due to the highly imbalanced Web-scale training set, these foundation models are inevitably skewed toward frequent semantics, and thus the subsequent fine-tuning or ensembling is still biased. In this study, we systematically examine the biases in foundation models and demonstrate the efficacy of our proposed Generalized Logit Adjustment (GLA) method. Note that bias estimation in foundation models is challenging, as most pre-train data cannot be explicitly accessed like in traditional long-tailed classification tasks. To this end, GLA has an optimization-based bias estimation approach for debiasing foundation models. As our work resolves a fundamental flaw in the pre-training, the proposed GLA demonstrates significant improvements across a diverse range of tasks: it achieves 1.5 pp accuracy gains on ImageNet, an large average improvement (1.4-4.6 pp) on 11 few-shot datasets, 2.4 pp gains on long-tailed classification. Codes are in https://github.com/BeierZhu/GLA.