TY  - RPRT
TI  - An initial alignment between neural network and target is needed for gradient descent to learn
AU  - Emmanuel Abbe
AU  - Elisabetta Cornacchia
AU  - Jan Hązła
AU  - Christopher Marquis
PY  - 2022
UR  - https://arxiv.org/abs/2202.12846
ID  - 2202.12846
ER  -