Codebook Features: Sparse and Discrete Interpretability for Neural Networks.

Alex Tamkin, Mohammad Taufeeque, Noah D. Goodman

Browse the full ICML paper archive.