Predicting the generalization gap in neural networks using topological data analysis

Understanding how neural networks generalize on unseen data is crucial for designing more robust and reliable models. In this paper, we study the generalization gap of neural networks using methods from topological data analysis. For this purpose, we compute homological persistence diagrams of weigh...

Descripción completa

Detalles Bibliográficos
Autores: Ballester Bautista, Rubén, Arnal i Clemente, Xavier, Casacuberta, Carles, Madadi, Meysam, Corneanu, Ciprian Adrian, Escalera Guerrero, Sergio
Tipo de recurso: artículo
Estado:Versión publicada
Fecha de publicación:2024
País:España
Institución:Universidad de Barcelona
Repositorio:Dipòsit Digital de la UB
OAI Identifier:oai:diposit.ub.edu:2445/218045
Acceso en línea:https://hdl.handle.net/2445/218045
Access Level:acceso abierto
Palabra clave:Topologia
Aprenentatge automàtic
Xarxes neuronals (Informàtica)
Topology
Machine learning
Neural networks (Computer science)
Descripción
Sumario:Understanding how neural networks generalize on unseen data is crucial for designing more robust and reliable models. In this paper, we study the generalization gap of neural networks using methods from topological data analysis. For this purpose, we compute homological persistence diagrams of weighted graphs constructed from neuron activation correlations after a training phase, aiming to capture patterns that are linked to the generalization capacity of the network. We compare the usefulness of different numerical summaries from persistence diagrams and show that a combination of some of them can accurately predict and partially explain the generalization gap without the need of a test set. Evaluation on two computer vision recognition tasks (CIFAR10 and SVHN) shows competitive generalization gap prediction when compared against state-of-the-art methods.