Towards Principled Evaluations of Sparse Autoencoders for Interpretability and Control.

Aleksandar Makelov, Georg Lange, Neel Nanda

Browse the full ICLR paper archive.