Striving for data-model efficiency: Identifying data externalities on group performance

TL;DR

This paper investigates how adding training data can sometimes harm subgroup performance in machine learning, highlighting the importance of understanding data-model interactions for building trustworthy AI.

Contribution

It introduces the concept of negative data externalities on group performance and analyzes how data and model size influence these externalities.

Findings

Data externalities can reduce subgroup performance when adding data.

Externalities vary with training set size and model complexity.

Understanding externalities is crucial for improving data efficiency.

Abstract

Building trustworthy, effective, and responsible machine learning systems hinges on understanding how differences in training data and modeling decisions interact to impact predictive performance. In this work, we seek to better understand how we might characterize, detect, and design for data-model synergies. We focus on a particular type of data-model inefficiency, in which adding training data from some sources can actually lower performance evaluated on key sub-groups of the population, a phenomenon we refer to as negative data externalities on group performance. Such externalities can arise in standard learning settings and can manifest differently depending on conditions between training set size and model size. Data externalities directly imply a lower bound on feasible model improvements, yet improving models efficiently requires understanding the underlying data-model tensions. From a broader perspective, our results indicate that data-efficiency is a key component of both accurate and trustworthy machine learning.