ITP: Instance-Aware Test Pruning for Out-of-Distribution Detection

TL;DR

This paper introduces ITP, a novel post-hoc method for out-of-distribution detection that adaptively prunes parameters at both coarse and fine levels to improve reliability and reduce overconfidence.

Contribution

ITP is a new instance-aware test pruning approach that enhances OOD detection by adaptively removing overconfident parameters based on contribution distributions and statistical tests.

Findings

ITP outperforms existing methods on standard benchmarks.

ITP effectively reduces overconfidence in OOD detection.

ITP demonstrates competitive or superior performance in experiments.

Abstract

Out-of-distribution (OOD) detection is crucial for ensuring the reliable deployment of deep models in real-world scenarios. Recently, from the perspective of over-parameterization, a series of methods leveraging weight sparsification techniques have shown promising performance. These methods typically focus on selecting important parameters for in-distribution (ID) data to reduce the negative impact of redundant parameters on OOD detection. However, we empirically find that these selected parameters may behave overconfidently toward OOD data and hurt OOD detection. To address this issue, we propose a simple yet effective post-hoc method called Instance-aware Test Pruning (ITP), which performs OOD detection by considering both coarse-grained and fine-grained levels of parameter pruning. Specifically, ITP first estimates the class-specific parameter contribution distribution by exploring the ID data. By using the contribution distribution, ITP conducts coarse-grained pruning to eliminate redundant parameters. More importantly, ITP further adopts a fine-grained test pruning process based on the right-tailed Z-score test, which can adaptively remove instance-level overconfident parameters. Finally, ITP derives OOD scores from the pruned model to achieve more reliable predictions. Extensive experiments on widely adopted benchmarks verify the effectiveness of ITP, demonstrating its competitive performance.