Evidence-Guided Unknown Rejection for High-Confidence Near-Known Unknowns

2026-05-17

AI Summary

This work addresses the challenge in open-set recognition where "near-known unknown" samples are often misclassified with high confidence as known classes by closed-set classifiers. To mitigate this issue, the authors propose EGUR-A, a post-hoc method that redefines the open-set decision logic: rather than assessing whether a sample’s score is sufficiently high, it evaluates whether the predicted class possesses adequate evidence to accept the sample. EGUR-A integrates class-conditional local acceptance evidence with global residual evidence and adaptively weights them based on known-class statistics—without requiring any unknown validation data. Evaluated as a plug-in module for off-the-shelf closed-set classifiers, EGUR-A significantly reduces high-confidence false acceptance rates on benchmarks such as CUB, FGVC-Aircraft, and ImageNet-hard, and consistently outperforms existing threshold-based approaches at matched operating points for rejection.

Abstract

Open-set recognition systems face a neglected failure mode: high-confidence near-known unknowns, which lie outside the known label set but are close enough to known classes that a closed-set classifier accepts them with high confidence. We show that this failure is widespread across scalar-threshold methods, including recent post-hoc detectors, and that stronger encoders can amplify rather than remove the risk. We propose EGUR-A, which changes the decision from ``is this sample's score high enough?'' to ``does this predicted known class have sufficient evidence to accept this sample?'' EGUR-A combines class-conditional local acceptance evidence with global residual evidence, and selects their relative weight from known-sample statistics without unknown validation data. Across CUB, FGVC-Aircraft, and ImageNet-hard, EGUR-A substantially reduces high-confidence false known acceptance at matched known-rejection operating points. The result is not a stronger threshold; it is a different question: whether a known class is entitled to accept a sample.

Problem

• open-set recognition
• near-known unknowns
• high-confidence errors
• unknown rejection
• closed-set classifier

Innovation

• open-set recognition
• unknown rejection
• evidence-guided decision
• near-known unknowns
• class-conditional evidence