MANO: Exploiting Matrix Norm for Unsupervised Accuracy Estimation Under Distribution Shifts

TL;DR

This paper introduces MaNo, a novel method that normalizes logits and uses their matrix norm to accurately estimate neural network accuracy under distribution shifts without labels.

Contribution

MaNo is a new unsupervised accuracy estimation method that reduces bias through data-dependent normalization and leverages matrix norms, showing state-of-the-art results.

Findings

MaNo outperforms existing methods on various benchmarks.

It maintains high accuracy under synthetic, natural, and subpopulation shifts.

Theoretical analysis links the score to model uncertainty.

Abstract

Leveraging the models' outputs, specifically the logits, is a common approach to estimating the test accuracy of a pre-trained neural network on out-of-distribution (OOD) samples without requiring access to the corresponding ground truth labels. Despite their ease of implementation and computational efficiency, current logit-based methods are vulnerable to overconfidence issues, leading to prediction bias, especially under the natural shift. In this work, we first study the relationship between logits and generalization performance from the view of low-density separation assumption. Our findings motivate our proposed method MaNo which (1) applies a data-dependent normalization on the logits to reduce prediction bias, and (2) takes the $L_p$ norm of the matrix of normalized logits as the estimation score. Our theoretical analysis highlights the connection between the provided score and the model's uncertainty. We conduct an extensive empirical study on common unsupervised accuracy estimation benchmarks and demonstrate that MaNo achieves state-of-the-art performance across various architectures in the presence of synthetic, natural, or subpopulation shifts. The code is available at \url{https://github.com/Renchunzi-Xie/MaNo}.