Decomposing the dark matter of sparse autoencoders.
Joshua Engels, Logan Riggs, Max Tegmark MIT 2024 https://arxiv.org/abs/2410.
On mapping concepts in artificial neural networks with sparse autoencoders: we find that map errors exhibit…
Code for our paper ‘Decomposing The Dark Matter of Sparse Autoencoders’ — JoshEngels/SAE-Dark-Matter.
Leave a reply