Towards Fairness in Visual Recognition: Effective Strategies for Bias Mitigation

TL;DR

This paper introduces a new benchmark for bias mitigation in visual recognition, systematically compares existing techniques, and proposes a simple, effective method that outperforms others in reducing gender bias.

Contribution

It provides a comprehensive analysis of bias mitigation strategies, highlights limitations of adversarial training, and introduces a novel domain-independent training technique.

Findings

The proposed technique outperforms existing methods in bias mitigation.

Adversarial training approaches have notable shortcomings.

The new method effectively reduces gender bias in CelebA attribute classification.

Abstract

Computer vision models learn to perform a task by capturing relevant statistics from training data. It has been shown that models learn spurious age, gender, and race correlations when trained for seemingly unrelated tasks like activity recognition or image captioning. Various mitigation techniques have been presented to prevent models from utilizing or learning such biases. However, there has been little systematic comparison between these techniques. We design a simple but surprisingly effective visual recognition benchmark for studying bias mitigation. Using this benchmark, we provide a thorough analysis of a wide range of techniques. We highlight the shortcomings of popular adversarial training approaches for bias mitigation, propose a simple but similarly effective alternative to the inference-time Reducing Bias Amplification method of Zhao et al., and design a domain-independent training technique that outperforms all other methods. Finally, we validate our findings on the attribute classification task in the CelebA dataset, where attribute presence is known to be correlated with the gender of people in the image, and demonstrate that the proposed technique is effective at mitigating real-world gender bias.