Underspecification Presents Challenges for Credibility in Modern Machine Learning

TL;DR

This paper highlights how underspecification in ML pipelines causes unpredictable and poor real-world performance, emphasizing the need to address this issue for reliable deployment across various domains.

Contribution

It identifies underspecification as a key factor behind deployment failures and demonstrates its prevalence and impact across multiple practical ML applications.

Findings

Underspecification leads to diverse predictor behaviors in deployment.

Models with similar training performance can behave very differently in real-world settings.

Addressing underspecification is crucial for reliable ML deployment.

Abstract

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.