DCAST: Diverse Class-Aware Self-Training Mitigates Selection Bias for Fairer Learning

TL;DR

This paper introduces DCAST, a novel self-training method that promotes sample diversity and addresses class-specific and hierarchy biases, leading to fairer and more robust models across various datasets.

Contribution

DCAST is a model-agnostic approach that mitigates class-specific bias and promotes diversity, improving fairness and robustness in machine learning models.

Findings

DCAST outperforms conventional self-training and domain adaptation methods on eleven datasets.

DCAST shows the largest advantage in multi-class classification tasks.

Models trained with DCAST demonstrate increased robustness to hierarchy and other biases.

Abstract

Fairness in machine learning seeks to mitigate model bias against individuals based on sensitive features such as sex or age, often caused by an uneven representation of the population in the training data due to selection bias. Notably, bias unascribed to sensitive features is challenging to identify and typically goes undiagnosed, despite its prominence in complex high-dimensional data from fields like computer vision and molecular biomedicine. Strategies to mitigate unidentified bias and evaluate mitigation methods are crucially needed, yet remain underexplored. We introduce: (i) Diverse Class-Aware Self-Training (DCAST), model-agnostic mitigation aware of class-specific bias, which promotes sample diversity to counter confirmation bias of conventional self-training while leveraging unlabeled samples for an improved representation of the underlying population; (ii) hierarchy bias, multivariate and class-aware bias induction without prior knowledge. Models learned with DCAST showed improved robustness to hierarchy and other biases across eleven datasets, against conventional self-training and six prominent domain adaptation techniques. Advantage was largest on multi-class classification, emphasizing DCAST as a promising strategy for fairer learning in different contexts.