Universal Sparse Autoencoders: Interpretable Cross-Model Concept Alignment.

Harrish Thasarathan, Julian Forsyth, Thomas Fel, Matthew Kowal, Konstantinos G. Derpanis

Browse the full ICML paper archive.