Paper

Can SAEs Capture Neural Geometry?

Bhalla, Geiger, Fel, Lubana, Rager, Feucht, Haklay, Wurgaft, Boppana, Kowal, Shyam, Lewis, McGrath, Merullo · 2026

Read the paper →

How sparse autoencoders represent (or fail to represent) curved geometric structures in neural networks, surfacing three modes: shattering, compact capture, and dilution.
mech-interpinterpretabilitymanifoldssparse-autoencoders