TY  - RPRT
TI  - Feature learning as alignment: a structural property of gradient descent in non-linear neural networks
AU  - Daniel Beaglehole
AU  - Ioannis Mitliagkas
AU  - Atish Agarwala
PY  - 2024
UR  - https://arxiv.org/abs/2402.05271
ID  - 2402.05271
ER  -