We gratefully acknowledge support from
the Simons Foundation and member institutions.
Full-text links:


Current browse context:


Change to browse by:

References & Citations

DBLP - CS Bibliography


(what is this?)
CiteULike logo BibSonomy logo Mendeley logo del.icio.us logo Digg logo Reddit logo ScienceWISE logo

Computer Science > Machine Learning

Title: Phase Diagram of Initial Condensation for Two-layer Neural Networks

Abstract: The phenomenon of distinct behaviors exhibited by neural networks under varying scales of initialization remains an enigma in deep learning research. In this paper, based on the earlier work by Luo et al.~\cite{luo2021phase}, we present a phase diagram of initial condensation for two-layer neural networks. Condensation is a phenomenon wherein the weight vectors of neural networks concentrate on isolated orientations during the training process, and it is a feature in non-linear learning process that enables neural networks to possess better generalization abilities. Our phase diagram serves to provide a comprehensive understanding of the dynamical regimes of neural networks and their dependence on the choice of hyperparameters related to initialization. Furthermore, we demonstrate in detail the underlying mechanisms by which small initialization leads to condensation at the initial training stage.
Subjects: Machine Learning (cs.LG); Disordered Systems and Neural Networks (cond-mat.dis-nn); Optimization and Control (math.OC); Machine Learning (stat.ML)
MSC classes: 68U99, 90C26, 34A45
Cite as: arXiv:2303.06561 [cs.LG]
  (or arXiv:2303.06561v2 [cs.LG] for this version)

Submission history

From: Yuqing Li [view email]
[v1] Sun, 12 Mar 2023 03:55:38 GMT (113kb,D)
[v2] Sat, 8 Apr 2023 00:12:27 GMT (841kb,D)

Link back to: arXiv, form interface, contact.