arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2209.00746cs.LGcs.CV
收藏

MIME:少数群体参与,多数群体参与,提高AI绩效

MIME: Minority Inclusion for Majority Group Enhancement of AI Performance

UCLA(加州大学洛杉矶分校)

Pradyumna Chari, Yunhao Ba, Shreeram Athreya, Achuta Kadambi

摘要:

有几篇论文正确地将少数群体纳入人工智能(AI)训练数据中,以改善少数群体和/或整个社会的测试推理。整个社会由少数和多数利益攸关方组成。一个常见的误解是,少数群体的纳入并不能单独提高多数群体的业绩。在本文中,我们做出了令人惊讶的发现,包括少数样本可以改善大多数群体的测试误差。换句话说,少数群体的包容性导致了多数群体的性能增强(MIME)。MIME效应的理论存在性证明,并发现与实验结果在六个不同的数据集是一致的。项目网页:https://visual.ee.ucla.edu/mime.htm/

英文摘要:

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/

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑