Predicting Out-of-Domain Generalization with Neighborhood Invariance

TL;DR

This paper introduces neighborhood invariance, a simple, assumption-free measure to predict out-of-domain generalization of classifiers by assessing their output stability under local transformations, applicable across various tasks and models.

Contribution

It proposes a novel, easy-to-compute measure of classifier invariance that does not rely on distributional assumptions or model gradients, enabling out-of-domain generalization prediction.

Findings

Strong correlation between neighborhood invariance and OOD generalization

Applicable across image, text classification, and natural language inference

Effective on over 4,600 models and 100 domain pairs

Abstract

Developing and deploying machine learning models safely depends on the ability to characterize and compare their abilities to generalize to new environments. Although recent work has proposed a variety of methods that can directly predict or theoretically bound the generalization capacity of a model, they rely on strong assumptions such as matching train/test distributions and access to model gradients. In order to characterize generalization when these assumptions are not satisfied, we propose neighborhood invariance, a measure of a classifier's output invariance in a local transformation neighborhood. Specifically, we sample a set of transformations and given an input test point, calculate the invariance as the largest fraction of transformed points classified into the same class. Crucially, our measure is simple to calculate, does not depend on the test point's true label, makes no assumptions about the data distribution or model, and can be applied even in out-of-domain (OOD) settings where existing methods cannot, requiring only selecting a set of appropriate data transformations. In experiments on robustness benchmarks in image classification, sentiment analysis, and natural language inference, we demonstrate a strong and robust correlation between our neighborhood invariance measure and actual OOD generalization on over 4,600 models evaluated on over 100 unique train/test domain pairs.