TY  - RPRT
TI  - Trainable at Depth, Solvable at Small Size: What the Loss Surface of a Polynomial Activation Dictates for Designing, Initializing and Training Networks
AU  - Xiuyi Yang
PY  - 2026
UR  - https://arxiv.org/abs/2610.20069
ID  - 2610.20069
ER  -