Dr. Jamie A. O’Reilly
School of International & Interdisciplinary Engineering Programs (SIIE), KMITL
Title: Recurrent neural network models of event-related brain signals
Abstract
Computational models of evoked and event-related electromagnetic signals from the brain can help researchers to interpret the neurophysiology underlying sensory and cognitive processing. Recurrent neural networks (RNNs) are one type of model that can be used for studying neurophysiological signal generation. These artificial neural networks are designed for handling sequential data, and are capable of learning complex nonlinear transformations between input sequences and output label sequences. Representations of experimental events are constructed for inputs, and segments of electrophysiology recordings are used as output labels. Each input sequence can be paired with multiple output sequences that were recorded during the associated experimental events. By training the RNN to minimize mean-squared-error loss, this one-to-many mapping causes the model outputs to converge towards the average of those output labels. After training the RNN, its behavior can be evaluated as a model for studying event-related brain signal generation. Simulations performed with RNN models can test and generate new predictions, thereby refining and formulating hypotheses about the neurological processes that produce these signals. In this talk, I will provide an overview of the RNN modelling approach and highlight recent developments in this area including (i) using RNN for blind source separation, (ii) modelling the effects of state on evoked potentials, (iii) simulating loudness dependence of the auditory response, (iv) modelling neural correlates of face perception, and (v) localized estimation of neural sources of evoked fields. This approach to modelling neural signals offers considerable flexibility, and the aforementioned examples merely scratch the surface of what could be possible using this technology. I will conclude the talk by summarizing some of the potential future directions for this program of research.
Autobiography
I come from a small village in Aberdeenshire, in the North-East of Scotland. In 2011, I graduated with a Master of Engineering (With Distinction) in Electronic and Electrical Engineering from Robert Gordon University, Aberdeen, then headed down to the South of Scotland for postgraduate studies. In 2018, I graduated with a Doctor of Engineering in Biomedical Engineering from the University of Strathclyde, Glasgow. My doctoral research involved conducting in-vivo electrophysiology experiments to investigate auditory processing deficits related to schizophrenia syndrome. After graduating, I was appointed as a Lecturer at the College of Biomedical Engineering, Rangsit University, Thailand. There I expanded my knowledge of machine learning, signal and image processing, and software engineering. In 2020, I was promoted to Assistant Professor of Biomedical Engineering at Rangsit University. In 2022, I briefly worked as a Visiting Scholar at Macquarie University, Australia, before joining the School of International and Interdisciplinary Engineering (SIIE) at King Mongkut’s Institute of Technology Ladkrabang (KMITL), Thailand. As an early-career researcher, I have predominantly contributed to the field of auditory neurophysiology research. Most relevant for the current talk, however, will be that I have recently spearheaded several methodological innovations incorporating the use of recurrent neural networks for event-related neural signal analysis.