TY  - RPRT
TI  - Saddle-to-Saddle Dynamics in Deep Linear Networks: Small Initialization Training, Symmetry, and Sparsity
AU  - Arthur Jacot
AU  - François Ged
AU  - Berfin Şimşek
AU  - Clément Hongler
AU  - Franck Gabriel
PY  - 2022
UR  - https://arxiv.org/abs/2106.15933
ID  - 2106.15933
ER  -