PRIME: Prioritizing Interpretability in Failure Mode Extraction

arXiv:2310.00164v2 Announce Type: replace
Abstract: In this work, we study the challenge of providing human-understandable descriptions for failure modes in trained image classification models. Existing works address this problem by first identifying clusters (or directions) of incorrectly classified samples in a latent space and then aiming to provide human-understandable text descriptions for them. We observe that in some cases, describing text does not match well with identified failure modes, partially owing to the fact that shared interpretable attributes of failure …

abstract arxiv challenge classification cs.cv extraction failure human image interpretability prime samples space study text them type work