Underspecification Presents Challenges for Credibility in Modern Machine Learning

作者：A. D'Amour, K. Heller, D. Moldovan, Ben Adlam, Babak Alipanahi, Alex Beutel, Christina Chen, Jonathan Deaton, Jacob Eisenstein, M. Hoffman, F. Hormozdiari, N. Houlsby, Shaobo Hou, Ghassen Jerfel, A. Karthikesalingam, M. Lučić, Yi-An Ma, Cory Y. McLean, Diana Mincu, Akinori Mitani, A. Montanari, Zachary Nado, Vivek Natarajan, Christopher Nielson, Thomas F. Osborne, R. Raman, K. Ramasamy, R. Sayres, Jessica Schrouff, Martin G. Seneviratne, Shannon Sequeira, H. Suresh, Victor Veitch, Max Vladymyrov, Xuezhi Wang, Kellie Webster, Steve Yadlowsky, T. Yun, Xiaohua Zhai, D. Sculley · 发表于：Journal of machine learning research · 年份：2020 · 被引用次数：939 · 研究领域：Computer Science、Mathematics

ML models often exhibit unexpectedly poor behavior when they are deployed in real-world domains. We identify underspecification as a key reason for these failures. An ML pipeline is underspecified when it can return many predictors with equivalently strong held-out performance in the training domain. Underspecification is common in modern ML pipelines, such as those based on deep learning. Predictors returned by underspecified pipelines are often treated as equivalent based on their training domain performance, but we show here that such predictors can behave very differently in deployment domains. This ambiguity can lead to instability and poor model behavior in practice, and is a distinct failure mode from previously identified issues arising from structural mismatch between training and deployment domains. We show that this problem appears in a wide variety of practical ML pipelines, using examples from computer vision, medical imaging, natural language processing, clinical risk prediction based on electronic health records, and medical genomics. Our results show the need to explicitly account for underspecification in modeling pipelines that are intended for real-world deployment in any domain.