TY  - RPRT
TI  - Backward Feature Correction: How Deep Learning Performs Deep (Hierarchical) Learning
AU  - Zeyuan Allen-Zhu
AU  - Yuanzhi Li
PY  - 2023
UR  - https://arxiv.org/abs/2001.04413
ID  - 2001.04413
ER  -