Papers
Bilinear Decomposition
(Jun 2024)Using bilinear MLPs to reverse-engineer shallow MNIST and Tiny Stories models using only their weights.
Tokenized SAEs
(Jun 2024)We propose using a per-token bias in SAEs to separate token reconstructions from interesting, semantic features.
The Trifecta
(Nov 2023)Three techniques to significantly improve the Forward-Forward algorithm. We achieve 84% on CIFAR-10.
And much more to come...