ConceptMechanistic Interpretability / 机制解释性
Mechanistic Interpretability
A research paradigm in AI explainability that aims to develop a deep understanding of the internal representational mechanisms and computational processes of neural networks, rather than merely characterizing their input-output behavior.
Timeline (last 90 days)
Sep 11
Mechanistic Interpretability is a research paradigm that attempts to open the black box and understand the internal representational mechanisms of models, and is receiving increasing attention in both academia and industry.
Unverified50%