差別データ多様性を意識した敵対的標本に基づく公平性テスト算法

抄録

In recent years, decision-making algorithms based on machine learning have become widely used in everyday life. While these algorithms often make more accurate judgments than humans, they can also learn biases and potentially compromise fairness. This study proposes an enhancement to an existing method that uses deep neural networks (DNN) to search for discriminatory data. By incorporating diversity into the search process, we demonstrate that our method can more effectively identify a wider range of discriminatory data compared to traditional approaches.