Study of a neuroevolution-based approach for nonlinear blind source separation

• Grant number: 21/11723-0
• Support Opportunities: Scholarships abroad - Research Internship - Scientific Initiation
• Start date: March 01, 2022
• End date: May 31, 2022
• Field of knowledge: Physical Sciences and Mathematics - Computer Science - Computer Systems
• Principal Investigator: Denis Gustavo Fantinato
• Grantee: Lucas Fernandes Muniz
• Supervisor: Christian Jutten
• Host Institution: Centro de Matemática, Computação e Cognição (CMCC). Universidade Federal do ABC (UFABC). Ministério da Educação (Brasil). Santo André , SP, Brazil
• Institution abroad: Université Grenoble Alpes (UGA), France
• Associated to the scholarship: 20/16456-7 - Study of neuroevolutive strategies for training and topological adaptation of artificial neural networks, BP.IC

Abstract

The Blind Source Separation problem is a model that can be applied to solve several real-world problems that are modeled as a mixture of independent components. Unlike its linear case, a general solution framework for the Nonlinear Blind Source Separation (NLBSS) problem has not been achieved yet. Some approaches based on constrained models have been proposed so far, however, the success of some models depends on the choice of an adequate order of complexity for the model. Based on it, the objective of this project is to solve the NLBSS problem using nonlinear Generalized Additive Models with adequate complexity. A suitable model will be achieved by using the NeuroEvolution of Augmenting Topologies algorithm, which is usually used to search for artificial neural networks. The statistical independence shall be evaluated by a mutual information rate that incorporates temporal information, in order to provide additional information to perform nonlinear source separation. (AU)