TY  - RPRT
TI  - A "feature ODE" describing the learning behavior of shallow MLPs on simple functions
AU  - Joseph Turnbull
AU  - Berkan Ottlik
AU  - Jamie Simon
PY  - 2026
UR  - https://arxiv.org/abs/2610.25652
ID  - 2610.25652
ER  -