The loss surfaces of neural networks with general activation functions. (1st June 2021)

• Record Type: Journal Article
• Authors: Baskerville, Nicholas P; Keating, Jonathan P; Mezzadri, Francesco; Najnudel, Joseph
• Is Part Of: Journal of statistical mechanics. (2021:Jun.)
• Publication Date: 2021-06-01
• Subjects: deep learning -- random matrix theory and extensions -- machine learning -- spin glasses; Statistical mechanics -- Periodicals; Mechanics -- Statistical methods -- Periodicals; 530.1305
• DOI: 10.1088/1742-5468/abfa1e
• Languages: English
• ISSNs: 1742-5468

Abstract

Abstract: The loss surfaces of deep neural networks have been the subject of several studies, theoretical and experimental, over the last few years. One strand of work considers the complexity, in the sense of local optima, of high dimensional random functions with the aim of informing how local optimisation methods may perform in such complicated settings. Prior work of Choromanska et al (2015) established a direct link between the training loss surfaces of deep multi-layer perceptron networks and spherical multi-spin glass models under some very strong assumptions on the network and its data. In this work, we test the validity of this approach by removing the undesirable restriction to ReLU activation functions. In doing so, we chart a new path through the spin glass complexity calculations using supersymmetric methods in random matrix theory which may prove useful in other contexts. Our results shed new light on both the strengths and the weaknesses of spin glass models in this context.