Evaluating Fairness in Self-supervised and Supervised Models for Sequential Data

2024/01/03 by Sofia Yfantidou, Dimitris Spathis, Yfantidou, Sofia +9
Social Sciences · #Computers and Society (cs.CY) #FOS: Computer and information sciences #Machine Learning (cs.LG) #Technology Use by Older Adults

paper · pdf · doi:10.48550/arxiv.2401.01640

openalex publication_date 2024/01/03 · openalex created_date 2024/01/05 · openalex updated_date 2026/08/04

Abstract

Self-supervised learning (SSL) has become the de facto training paradigm of large models where pre-training is followed by supervised fine-tuning using domain-specific data and labels. Hypothesizing that SSL models would learn more generic, hence less biased, representations, this study explores the impact of pre-training and fine-tuning strategies on fairness (i.e., performing equally on different demographic breakdowns). Motivated by human-centric applications on real-world timeseries data, we interpret inductive biases on the model, layer, and metric levels by systematically comparing SSL models to their supervised counterparts. Our findings demonstrate that SSL has the capacity to achieve performance on par with supervised methods while significantly enhancing fairness--exhibiting up to a 27% increase in fairness with a mere 1% loss in performance through self-supervision. Ultimately, this work underscores SSL's potential in human-centric computing, particularly high-stakes, data-scarce application domains like healthcare.

Related

• Using Self-supervised Learning Can Improve Model Fairness
• Evaluating the fairness of fine-tuning strategies in self-supervised learning
• Occam's Razor for Self Supervised Learning: What is Sufficient to Learn Good Representations?
• FAIRWELL: Fair Multimodal Self-Supervised Learning for Wellbeing Prediction
• On Pretraining Data Diversity for Self-Supervised Learning