English

Neural Collapse with Cross-Entropy Loss

Machine Learning 2021-01-20 v2 Classical Analysis and ODEs

Abstract

We consider the variational problem of cross-entropy loss with nn feature vectors on a unit hypersphere in Rd\mathbb{R}^d. We prove that when dn1d \geq n - 1, the global minimum is given by the simplex equiangular tight frame, which justifies the neural collapse behavior. We also prove that as nn \rightarrow \infty with fixed dd, the minimizing points will distribute uniformly on the hypersphere and show a connection with the frame potential of Benedetto & Fickus.

Cite

@article{arxiv.2012.08465,
  title  = {Neural Collapse with Cross-Entropy Loss},
  author = {Jianfeng Lu and Stefan Steinerberger},
  journal= {arXiv preprint arXiv:2012.08465},
  year   = {2021}
}
R2 v1 2026-06-23T20:59:35.746Z