MCCE builds pseudo-concepts from a model's hidden states, orthogonal to observed concepts, to reduce bias in causal concept effect estimates when some concepts are unobserved.
Machine intelligence in healthcare—perspectives on trustworthiness, explainability, usability, and transparency
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2024 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
MCCE: Missingness-aware Causal Concept Explainer
MCCE builds pseudo-concepts from a model's hidden states, orthogonal to observed concepts, to reduce bias in causal concept effect estimates when some concepts are unobserved.