Skip to main content

Neural Network : An Overparametrization Perspective


Project description

More is better but why: understanding successful neural network training models

Neural networks can ‘learn’ from input data and scenarios, improving their predictive ability for similar and different problems with successive iterations. So-called overparameterised models are among the most popular for training neural networks. These have more parameters than can be estimated from the training data, i.e. there are more parameters than needed to perfectly fit all the data points. Despite their empirical success, theoretical understanding of how these models are optimised and how their results are generalised to yield universal approximation remains poorly understood. With the support of the Marie Skłodowska-Curie Actions programme, the NN-OVEROPT project will enhance understanding to provide better optimisation algorithms for training.

Coordinator

INSTITUT NATIONAL DE RECHERCHE EN INFORMATIQUE ET AUTOMATIQUE
Net EU contribution
€ 257 619,84
Address
Domaine De Voluceau Rocquencourt
78153 Le Chesnay Cedex
France

See on map

Region
Ile-de-France Ile-de-France Yvelines
Activity type
Research Organisations
Non-EU contribution
€ 0,00

Partners (1)

Partner

Partner organisations contribute to the implementation of the action, but do not sign the Grant Agreement.

THE BOARD OF TRUSTEES OF THE UNIVERSITY OF ILLINOIS
United States
Net EU contribution
€ 0,00
Address
506 S. Wright Street, 209 Hab, Mc 339
61801 Urbana Il

See on map

Activity type
Higher or Secondary Education Establishments
Non-EU contribution
€ 165 265,92