ECE Colloquium Series: Vahid Tarokh: Robust Brain Signal Analysis—Single and Cross Subject Studies
Phillips Hall 233
In this talk, we will first discuss eye movement decoding in a working memory experiment involving two macaque monkeys. Our first objective is to use the local field potentials (LFPs) collected from the brain of each monkey to decode the type of task that the same monkey is doing, and the direction of saccade in each task.
We will first show that the LFP time-series data can be modeled using a nonparametric regression framework, and show that the classifiers trained using minimax function estimators as features are robust and consistent. We will also discuss application of the resulting classifier to the brain data.
Next we will discuss our cross-monkey knowledge transfer, where we use what we learn from a first macaque monkey data to improve decoding of actions of a second macaque monkey.
Finally, we create new robust neural networks based on robust estimation ideas in order for application in limited data cases. We apply our constructions to and achieve decoding accuracy of up to 97% at superficial cortical depths and up to 70% at deep cortical sites, making an improvement of about 350% over the existing methods.
This talk is based on various joint papers with Taposh Banerjee, Marko Angjelichinoski, Bijan Pesaran, and John Choi.
Vahid Tarokh is a Professor of Electrical and Computer Engineering at Duke University. He has supervised 35 Post-doctoral Fellow and 16 Ph.D. students; more than 55% of these are Professors at Research Universities, and the rest are research scientists at various US Government sponsored (Lincoln Labs, NASA), and Industry Research Labs. Additionally, he has supervised 12 M.S. thesis, one undergraduate thesis, and three M.S. non-thesis students. In the summer of 2016, he supervised 6 (mainly underrepresented) High School Summer Student Research on development of tactile gloves and applications (while volunteering under a United States Army Education Outreach HSAP Program), and in the summer of 2019, he supervised the research of 3 Duke Undergraduate students on sentiment detection from human speech (under Rhodes Information Initiative at Duke DATA+ program). He received his Ph.D. degree in Electrical Engineering from the University of Waterloo, Ontario, Canada, in 1995.