TY  - RPRT
TI  - Geometry of learning dynamics: Gradient descent versus natural gradient on the ridge of optimization
AU  - Akira Tamamori
PY  - 2026
UR  - https://arxiv.org/abs/2609.16805
ID  - 2609.16805
ER  -