Fixes That Fail: Self-Defeating Improvements in Machine-Learning Systems

TL;DR

This paper demonstrates that improving individual machine-learning models within complex systems can sometimes worsen overall system performance due to model entanglement, highlighting the need for holistic evaluation.

Contribution

It introduces the concept of self-defeating improvements in ML systems, analyzes error types causing this, and provides experimental evidence in a stereo detection system.

Findings

Self-improving models can degrade overall system performance.

Error decomposition reveals error types leading to self-defeating improvements.

Experiments confirm the phenomenon in a real stereo detection system.

Abstract

Machine-learning systems such as self-driving cars or virtual assistants are composed of a large number of machine-learning models that recognize image content, transcribe speech, analyze natural language, infer preferences, rank options, etc. Models in these systems are often developed and trained independently, which raises an obvious concern: Can improving a machine-learning model make the overall system worse? We answer this question affirmatively by showing that improving a model can deteriorate the performance of downstream models, even after those downstream models are retrained. Such self-defeating improvements are the result of entanglement between the models in the system. We perform an error decomposition of systems with multiple machine-learning models, which sheds light on the types of errors that can lead to self-defeating improvements. We also present the results of experiments which show that self-defeating improvements emerge in a realistic stereo-based detection system for cars and pedestrians.