Neural networks: Modeling, Simulation, Optimization and Applications

Abstract

In this talk will be described the two main used architectures of neural networks (NN): networks of perceptrons, and spiking neurons. These networks are used to solve classification and regression problems. These NN are simulated with open source software with Python using the modules Science Kit Learn or Pytorch. The main problem with NN is that one cannot know in advance which is the number of hidden layers and the number of neurons in each hidden layer to solve a given application. I will present one Network Search Architecture proposed to optimize recurrent networks made with Echo State Networks. For each described architecture will be shown applications with simulated and real data.  

Bio: 

Dr. Luis Gerardo de la Fraga received the B.S. degree in electrical engineering from Instituto Tecnológico de Veracruz, México in 1992; the M.Sc degree from Instituto Nacional de Astrofísica Optica y Electrónica (INAOE), Puebla, México, in 1994; and the Ph.D. degree from the Autonomous University of Madrid, Spain, in 1998. Since 2000, he has been with the Computer Science Department at the Center of Research and Advanced Studies (Cinvestav) in Mexico City. His research areas include computer vision, optimization, machine learning, and network security. He is very enthusiastic about open software and GNU/Linux systems.

Dr. De la Fraga has published more than 100 articles in journals and international conferences, and 3 books. He has graduated 4 PhD and 32 MSc students. He belongs to SNII level II, and is a member of the Mexican Academy of Science, ACM, and IEEE societies since 2005.