Debiasing Classifiers by Amplifying Bias with Latent Diffusion and Large Language Models

TL;DR

DiffuBias is a novel method that uses pretrained diffusion models to generate bias-conflict samples, improving classifier robustness and generalization without additional training, and achieves state-of-the-art results in debiasing tasks.

Contribution

This paper introduces DiffuBias, the first approach leveraging stable diffusion models to generate bias-conflict samples for debiasing classifiers.

Findings

Achieves state-of-the-art debiasing performance on benchmark datasets.

Effectively enhances classifier robustness without additional training.

Demonstrates computational efficiency through energy consumption analysis.

Abstract

Neural networks struggle with image classification when biases are learned and misleads correlations, affecting their generalization and performance. Previous methods require attribute labels (e.g. background, color) or utilizes Generative Adversarial Networks (GANs) to mitigate biases. We introduce DiffuBias, a novel pipeline for text-to-image generation that enhances classifier robustness by generating bias-conflict samples, without requiring training during the generation phase. Utilizing pretrained diffusion and image captioning models, DiffuBias generates images that challenge the biases of classifiers, using the top-$K$ losses from a biased classifier ($f_B$) to create more representative data samples. This method not only debiases effectively but also boosts classifier generalization capabilities. To the best of our knowledge, DiffuBias is the first approach leveraging a stable diffusion model to generate bias-conflict samples in debiasing tasks. Our comprehensive experimental evaluations demonstrate that DiffuBias achieves state-of-the-art performance on benchmark datasets. We also conduct a comparative analysis of various generative models in terms of carbon emissions and energy consumption to highlight the significance of computational efficiency.