Debiasify: Self-Distillation for Unsupervised Bias Mitigation

TL;DR

Debiasify is a self-distillation method that unsupervisedly mitigates biases in neural networks, improving generalization and fairness across diverse datasets without requiring bias annotations.

Contribution

It introduces a novel unsupervised self-distillation technique that transfers knowledge within the network to learn debiased representations without prior bias knowledge.

Findings

Significantly improves worst-group accuracy in vision and medical imaging tasks.

Outperforms previous unsupervised debiasing methods by large margins.

Achieves comparable or better results than supervised approaches.

Abstract

Simplicity bias poses a significant challenge in neural networks, often leading models to favor simpler solutions and inadvertently learn decision rules influenced by spurious correlations. This results in biased models with diminished generalizability. While many current approaches depend on human supervision, obtaining annotations for various bias attributes is often impractical. To address this, we introduce Debiasify, a novel self-distillation approach that requires no prior knowledge about the nature of biases. Our method leverages a new distillation loss to transfer knowledge within the network, from deeper layers containing complex, highly-predictive features to shallower layers with simpler, attribute-conditioned features in an unsupervised manner. This enables Debiasify to learn robust, debiased representations that generalize effectively across diverse biases and datasets, improving both worst-group performance and overall accuracy. Extensive experiments on computer vision and medical imaging benchmarks demonstrate the effectiveness of our approach, significantly outperforming previous unsupervised debiasing methods (e.g., a 10.13% improvement in worst-group accuracy for Wavy Hair classification in CelebA) and achieving comparable or superior performance to supervised approaches. Our code is publicly available at the following link: Debiasify.