TULiP: Test-time Uncertainty Estimation via Linearization and Weight Perturbation

TL;DR

TULiP is a post-hoc uncertainty estimation method for out-of-distribution detection that leverages linearization and weight perturbation to provide reliable uncertainty scores, achieving state-of-the-art results.

Contribution

It introduces a theoretically-driven approach to estimate uncertainty by analyzing network perturbations based on linearized training dynamics, applicable post-training.

Findings

Achieves state-of-the-art OOD detection performance.

Effective for near-distribution samples.

Visualized bounds on synthetic datasets.

Abstract

A reliable uncertainty estimation method is the foundation of many modern out-of-distribution (OOD) detectors, which are critical for safe deployments of deep learning models in the open world. In this work, we propose TULiP, a theoretically-driven post-hoc uncertainty estimator for OOD detection. Our approach considers a hypothetical perturbation applied to the network before convergence. Based on linearized training dynamics, we bound the effect of such perturbation, resulting in an uncertainty score computable by perturbing model parameters. Ultimately, our approach computes uncertainty from a set of sampled predictions. We visualize our bound on synthetic regression and classification datasets. Furthermore, we demonstrate the effectiveness of TULiP using large-scale OOD detection benchmarks for image classification. Our method exhibits state-of-the-art performance, particularly for near-distribution samples.