Seminar July 04TH, 2023
Hours: 16:30 - 18:00
Carlos Valle Araya
Title: Towards natural speech decoding with non-invasive Brain-Machine Interfaces
Abstract: Brain-Computer Interfaces (BCIs) have revolutionized communication by translating brain activity into commands for interacting with the external world. In recent years, there has been a growing focus on non-invasive Electroencephalography (EEG)-based speech BCIs that aim to decode perceived and silent speech from language-related brain areas, such as Broca's and Wernicke's areas. These speech BCIs aim to decode speech directly from neural data and have the potential to enable more natural communication compared to conventional methods like typing individual characters (i.e., the use of the P300 signal).
Both speech and neural data exhibit complex non-linear dynamics, making linear decoding algorithms suboptimal for these tasks. Deep Neural Networks (DNNs) offer a solution to this challenge, due to the non-linear properties of the activation functions in most layers. However, DNNs require a large amount of data to be train which can be time consuming and expensive. We investigate the creation of decoding models with the ability to generalize across different subjects, thereby establishing universally applicable models that can efficiently perform with previously unseen individuals. By advancing these techniques, we aim to facilitate the practical implementation of speech decoding tools in real-world settings.
Bio: Carlos has a Bachelor's degree in Biomedical Engineering from the Pontifical Catholic University of Chile, Chile. He is a doctoral candidate in the Doctorate Program at the Institute of Biological and Medical Engineering (IIBM) of the Pontifical Catholic University of Chile under the guidance of Dr. MarC-a RodrC-guez-FernC!ndez and Dr. Carolina Mendez. His research aims to improve Brain-to-Text models that transform brain activity into textual representations in tasks of speech perception and internal monologue (silent speech) using Deep Learning and Transfer Learning. Furthermore, he is interested in the open-source community for the creation of tools that facilitate and enhance scientific research.
Alberto Di Biase Oemick
Title: Fast 3D cardiac T2 mapping with isotropic resolution enable by Deep Learning
Abstract: Parametric mapping of the heart can be useful for diagnosis of different myocardial disease. These work is based on the MUST-T2 sequence develop by Aurélien Bustin and collaborators at King's College London. This sequence is can do T2 mapping in 3D with isotropic resolution in less than 8 minutes. Originally, this sequence was reconstructed using a low rank patch based reconstruction (HD-PROST) and dictionary matching for mapping. This reconstruction method is slow and this work focus on accelerating reconstruction and mapping times using deep learning.
Bio: Alberto Di Biase has a master in electrical engineer from Pontificia Universidad Católica de Chile, with a focus in magnetic resonance imaging (MRI) and medical imaging. He has experience working on deep learning research to accelerate and improve MRI. Currently, he is working as a research engineer at the iHealth Millennium Institute for Intelligent Healthcare Engineer in Santiago, Chile.
