arXiv cs.LG
· Papers
Optimal generalisation and learning transition in extensive-width shallow neural networks near interpolation
arXiv:2501.18530v3 Announce Type: replace-cross Abstract: We consider a teacher-student model of supervised learning with a fully-trained two-layer neural network whose width $k$ and input dimension $d$ are large and proportional. We provide an effective theory for approximating the Bayes-optimal generalisation error o