TY  - RPRT
TI  - Doing the impossible: Why neural networks can be trained at all
AU  - Nathan O. Hodas
AU  - Panos Stinis
PY  - 2018
UR  - https://arxiv.org/abs/1805.04928
ID  - 1805.04928
ER  -