MIME: Minority Inclusion for Majority Group Enhancement of AI Performance

TL;DR

This paper demonstrates that including minority group data in AI training not only benefits minority groups but also enhances performance for majority groups, supported by theoretical proof and experiments across multiple datasets.

Contribution

It introduces the MIME effect, showing minority inclusion can improve majority group performance, with a theoretical proof and experimental validation.

Findings

Minority inclusion improves majority group test error.

The MIME effect is consistent across six datasets.

Theoretical proof supports experimental results.

Abstract

Several papers have rightly included minority groups in artificial intelligence (AI) training data to improve test inference for minority groups and/or society-at-large. A society-at-large consists of both minority and majority stakeholders. A common misconception is that minority inclusion does not increase performance for majority groups alone. In this paper, we make the surprising finding that including minority samples can improve test error for the majority group. In other words, minority group inclusion leads to majority group enhancements (MIME) in performance. A theoretical existence proof of the MIME effect is presented and found to be consistent with experimental results on six different datasets. Project webpage: https://visual.ee.ucla.edu/mime.htm/