Este site utiliza cookies para lhe proporcionar uma melhor experiência de utilização. Ao navegar aceita a política de cookies.

Talk no DEI com Jamal Toutouh

23 março
Talk no DEI
Talk no DEI
© Jamal Toutouh

Talk no DEI com Jamal Toutouh
23/03/2022 | 16:30 | Room A5.1

Robust and resilient GAN training through spatially distributed coevolution

In recent years, machine learning with Generative Adversarial Networks (GANs) has been recognized as a powerful method for generative modeling. GANs have been successfully used to generate realistic synthesized images (e.g., image super-resolution, photograph editing, and text-to-image translation), sound (e.g., voice translation and music generation), and video (e.g., video-to-video translation and AI-assisted video calls), finding application in domains of multimedia information, engineering, science, design, art, and games.

GANs are trained by applying an adversarial paradigm, in which two deep models compete with each other using antagonistic lost functions to train the parameters with gradient descent. This adversarial training often shows pathological situations that prevent ideal convergence (e.g., mode collapse, oscillation, and vanishing gradients). All of these common problems are areas of active research.

The arms race defined in GANs training connects them to natural evolution because it also exhibits adversarial engagements and competitive coevolution. Thus, different solutions have been proposed by the evolutionary computation (EC) community to offer means of solving convergence impasses often encountered in GAN training. During the talk, the main GANs pathologies will be introduced, and the EC (evolutionary and co-evolutionary) strategies provided by Lipizzaner will be described.

Jamal Toutouh is a Researcher Assistant Professor at the University of Málaga (Spain). Previously, he was a Marie Skłodowska Curie Postdoctoral Fellow at Massachusetts Institute of Technology (MIT) in the USA, at the MIT CSAIL Lab. He obtained his Ph.D. in Computer Engineering at the University of Malaga (Spain). During his doctoral studies, Jamal analyzed and devised Machine Learning and Optimization methods inspired by Nature to address Smart Mobility challenges in modern cities. His dissertation entitled "Natural Computing for Vehicular Networks" was awarded several prizes, including the Best Spanish Ph.D. Thesis in Smart Cities (2018).

Currently, he is a Generative Machine Learning enthusiast. His research explores the combination of Nature-inspired gradient-free approaches and gradient-based methods to address Deep Generative Modeling. Specifically, Jamal focuses on improving Generative Adversarial Networks (GAN) training effectiveness and efficiency in order to face different generative challenges related to Smart Mobility, Smart Cities, and Climate Change.