Universal characteristics of deep neural network loss surfaces from random matrix theory

Autor: Nicholas P Baskerville, Jonathan P Keating, Francesco Mezzadri, Joseph Najnudel, Diego Granziol

Publikováno v: Baskerville, N P, Keating, J, Mezzadri, F, Najnudel, J & Granziol, D 2022, ' Universal characteristics of deep neural network loss surfaces from random matrix theory ', Journal of Physics A: Mathematical and Theoretical, vol. 55, no. 49, 494002 . https://doi.org/10.1088/1751-8121/aca7f5

This paper considers several aspects of random matrix universality in deep neural networks. Motivated by recent experimental work, we use universal properties of random matrices related to local statistics to derive practical implications for deep ne

Externí odkaz: https://explore.openaire.eu/search/publication?articleId=doi_dedup___::ec082a1e277108d203a7bffc78ddaa8e
http://arxiv.org/abs/2205.08601

A Spin Glass Model for the Loss Surfaces of Generative Adversarial Networks

Autor: Nicholas P. Baskerville, Jonathan P. Keating, Francesco Mezzadri, Joseph Najnudel

Publikováno v: Journal of Statistical Physics
Baskerville, N P, Keating, J P, Mezzadri, F & Najnudel, J 2022, ' A spin-glass model for the loss surfaces of generative adversarial networks ', Journal of Statistical Physics, vol. 186, no. 2, 29 . https://doi.org/10.1007/s10955-022-02875-w
University of Bristol-PURE

We present a novel mathematical model that seeks to capture the key design feature of generative adversarial networks (GANs). Our model consists of two interacting spin glasses, and we conduct an extensive theoretical analysis of the complexity of th

Externí odkaz: https://explore.openaire.eu/search/publication?articleId=doi_dedup___::ca55a4d5482b4609bd67c6dc9a57ec9f
https://pubmed.ncbi.nlm.nih.gov/35125517