← back to paper
arxiv: 2605.07922 · 2 revisions
Tree SAE: Learning Hierarchical Feature Structures in Sparse Autoencoders