Latent Projections · Geometric Rotation · Sparsity Collapse

Publications

research contributions to geometric deep learning
and mechanistic interpretability

All Publications

Published

RT-TopKSAE: Improving Top-k Sparse Autoencoders with the Rotation Trick

Sulayman Yusuf, A. Balwani

ICLR 2026 @ Re-Align Research Workshop, 2026

Sparse AutoencodersGeometric Deep LearningMechanistic InterpretabilityRepresentation Learning

Riemannian Geometry · Gravity Wells · Orthographic Projection

Research Interests

each interest warps the manifold — well depth measures focus.
related fields share a valley in the geometry.