Interpretable Failure Detection with Human-Level Concepts

Kien X. Nguyen, Tang Li, Xi Peng

Reliable failure detection holds paramount importance in safety-critical applications. Yet, neural networks are known to produce overconfident predictions for misclassified samples…