Reducing sample needs for better model transfer across varied populations
Rural Readiness: Domain Generalization with Adversarially Chosen Domains
Machine Learning
Summary
Models trained in big hospitals often don’t work well in smaller rural ones due to differences in data. The authors studied how to build models that can perform well even when facing new, unseen groups by relaxing strict accuracy requirements slightly. They found this allows needing fewer examples from each group to train effective models. Also, having a small amount of data from a new group doesn’t necessarily help improve performance on difficult cases from previous groups.
What this means in practice
- •For healthcare data scientists: Design diagnostic models that better generalize from hospital data to diverse rural populations with fewer data samples needed per site.
- •For machine learning platform engineers: Improve domain generalization features in ML platforms by incorporating training methods that assume slight tolerance in source domain coverage for better efficiency.
A theory result. No direct application yet.
Authors
Cynthia Dwork, Andisheh Ghasemi, Lunjia Hu
Abstract
Predictive models trained at well-resourced data-rich institutions often fail when deployed in new populations. Motivated by the goal of building clinical diagnosis and decision-making algorithms for data-scarce rural populations using the general knowledge learned from data-rich hospitals, we formulate the fundamental question of \emph{domain generalization with adversarially chosen domains}, which asks a learner to perform on any unseen target domain nearly as well as the worst hypothesis consistent with the evidence from the observed source domains. Assuming only that the hypothesis class contains a "universal" predictor with small error on all $k$ source domains, we characterize the sample complexity of this objective up to logarithmic factors. Letting $d$ denote the VC dimension of the hypothesis class, we find, somewhat surprisingly, that the sample complexity reduces significantly from $Ω(dk)$ to $O(d)$ under a small relaxation of the learning goal in which, roughly speaking, we allow the learned model to compete with hypotheses that perform well on 99% instead of 100% of the source domains. By applying this insight to the meta-distributional setting, in which source domains are selected according to a distribution on distributions, we obtain a corresponding improvement over known sample complexity upper bounds. We also study the effect of data access to the target domain and show that a small labeled sample from the target domain does not improve the worst-case sample complexity on the source domains.