Empirical tests of VLM-CBMs show VLM supervision differs from expert annotations depending on task and that concept accuracy correlates weakly with quality metrics.
Evaluating the stability of semantic concept representations in CNNs for robust explainability
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2025 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
If Concept Bottlenecks are the Question, are Foundation Models the Answer?
Empirical tests of VLM-CBMs show VLM supervision differs from expert annotations depending on task and that concept accuracy correlates weakly with quality metrics.