Accepted tutorials

Nuevos modelos y algortimos de entrenamiento para redes neuronales artificiales.
Dr. Juan Humberto Sossa Azuela

Resumen:

En este tutorial se dará en primer lugar un panorama general sobre las redes neuronales artificiales (RNAS) basadas en neuronas) perceptrones de primera y segunda generación. Se hablará de sus ventajas y limitaciones. Enseguida, se exponrán detalles sobre RNAS basados en perceptrones morfológicos que incorporan procesamiento dendrítico. Se explicarán las ventajas de incorporar este tipo de procesamiento en la operación de las neuronas. Para cada nuevo modelo de red neuronal morfológica se expondrá la operación fundamental su algoritmo de entrenamiento. Se presentarán comparaciones contra modelos clásicos de RNAS. Finalmente, se presentarán propuestas de hibridaciones de RNAS y se hablará de sus ventajas y limitaciones sobre otros modelos al compararlos con otros modelos.

 ****************************************************************************************************

Brain-Computer Interface (BCI) and Machine Learning

Luis G. Hernandez, Ph.D student, Tecnologico de Monterrey
Carlos D. Virgilio G, Ph.D student, Instituto Politécnico Nacional
Javier M. Antelis, Professor-Researcher, Tecnologico de Monterrey
Juan H. Sossa, Professor-Researcher, Centro de Investigación en Computación-IPN

Resumen:

The last years have witnessed the maturity of Brain-Computer Interfaces (BCI). Basically, a BCI is an emerging technology that provides a new alternative to recover functionality of impaired limbs, therefore, they can be used as a new non-muscular communication channel for people with partial or complete motor disabilities. In this line, Tecnologico de Monterrey initiated a Neuroethology and Brain Computer Interfaces Laboratory (NTLab) which is a research facility sponsored by the National Council of Science and Technology of Mexico (CONACYT) that is focused to conduct research in BCI. Researchers at NTLab study the human brain electrical activity acquired with the non-invasive electroencephalogram (EEG) technique in order to develop novel solutions for BCI and seeks to contribute to the field of signal processing and computational intelligence for BCI.
A BCI relies on machine learning algorithms to detect patterns or changes in the ongoing EEG signals that encode a mental task performed by the user. This is a highly complex task because the EEG activity is nonstationary, presents strong variations across participants and sessions, lacks sufficient bandwidth to encode decipherable fine motor information, among others. These situations have been addressed with classical supervised learning approaches and several functional applications have been developed and tested with healthy and impaired persons. However, it is necessary to explore other classification models as those based on deep learning. In this line, novel bio-inspired classification algorithms as “Dendrite morphological neural networks” and “Deep Multi-Layer Perceptrons” have been developed at the “Centro de Investigación en Computación – Instituto Politécnico Nacional” which have been successfully applied in BCI systems.
This workshop aims to provide a practical experience into BCI and to show the importance of machine learning in these systems. Participants will work with a real EEG-based BCI system and they be able to control a computer cursor in two dimensions (left and right) by simply performing motor imagery tasks. To do so, they will analyze and visualize the recorded EEG signals and then they will apply feature extraction and supervised classification algorithms. From this workshop, it will naturally emerge a vivid discussion between professors and students where it will be identified the current technical and design challenges of BCIs in which the "computational intelligence" and "novel machine learning models" will be a key element to achieve BCI for real users and functional applications.

Methodology
1) Basis of BCIs and the role of machine learning. Oral presentation. 30mins
2) Practical basis of electroencephalogram recordings. Hands on work with an EEG system. Gathering EEG data for BCI calibration
    a. Hands on work an EEG system and a BCI SW platform
3) Data Analysis Part 1
    a. Data preprocessing
    b. Hands on work with Matlab/Python
4) Break
5) Data Analysis Part 2
    a. Feature extraction and selection
    b. Hands on work with Matlab/Python
6) Data Analysis Part 3
    a. Trainning and setting a classifier
    b. Hands on work with Matlab/Python
7) Online control using brain signals
    a. Hands on work with an EEG system

Topics to be covered

• Theoretical aspects and current state of BCIs
• Machine learning in BCIs
• Electroencephalography (EEG)
• Feature extraction and selection
• Supervised learning
• Performance metrics

*******************************************************************************************

Principios de Aprendizaje Profundo (Deep Learning) y su aplicación en reconocimiento de
expresiones faciales con TensorFlow-Keras.
Dr. Luis Eduardo Falcón Morales
Dr. Juan Humberto Sossa Azuela
Resumen:
En los últimos años las Redes Neuronales Profundas, Deep Learning o Aprendizaje Profundo, ha
estado teniendo un gran impacto y desarrollo en el mundo científico. En este tutorial abordaremos
la teoría que está detrás de estas redes neuronales, y la pondremos en práctica a través de
Keras+TensorFlow para el reconocimiento de expresiones faciales en imágenes o video.
Temas:
1. Principios de Redes Neuronales Artificiales
a. Aprendizaje supervisado
b. Regresión logística
c. Backpropagation
2. Introducción a las Redes Neuronales Convolucionales CNN
a. Arquitecturas
b. Operador convolución y capas de una CNN
3. Conociendo Keras+TensorFlow
a. Conceptos básicos Python+Keras+TensorFlow
b. Pre-procesamiento de las imágenes
4. Reconocimiento de expresiones faciales
a. Base de datos y entrenamiento
b. Validación de la red
Pre-requisitos:
Los participantes deben tener conocimientos de Python, además de que es deseable tener
conocimientos de Cálculo y Álgebra Matricial.
Speakers:
Luis Eduardo Falcón Morales
Juan Humberto Sossa Azuela
https://sites.google.com/site/cicvision/