← back to paper
arxiv: 2602.00924 · 2 revisions
Supervised sparse auto-encoders for interpretable and compositional representations