Title:  Towards Addressing Some Fundamental Challenges with Brain-Computer Interfaces: A Systems Approach

Committee: 

Dr. Sivakumar, Advisor 

Dr. Ji, Chair

Dr. Fekri

Abstract: The objective of the proposed research is to address some of the fundamental challenges in Brain-Computer Interface (BCI) systems. Brain-computer interfaces have surfaced as a powerful and attractive modality in human-machine interaction. BCIs establish a channel for communicating with machines directly, bypassing the physical constraints and limitations of the body. BCIs measure the electrical impulses inside the brain and use this information to indicate a user's intent to a machine. Non-invasive BCIs, which utilize a convenient device worn by a user are more popular than invasive BCIs which often require surgery. As of 2016, 71.2\% of all research on BCIs pertained to EEG-based non-invasive BCIs. However, EEG-based non-invasive BCIs also suffer from 2 fundamental challenges, among other problems. Firstly, since they measure the electrical activity in the brain externally on the scalp, they record a signal that has to pass through the skull, (thereby reducing its signal power) and which is a combination of the desired signal and signals from other parts of the brain, which significantly reduces the signal-to-noise ratio. This leads to low detection accuracy and poor performance by classification models. Secondly, BCI signals do not generalize well across other individuals owing to a high amount of signal variability between subjects. This makes it extremely difficult to re-use a machine learning model built by using one user's data on others and requires tedious re-training on new users. In this work, we attempt to address these problems using a systems approach. We view these challenges as subproblems in a wider system of brain-computer interfaces and their applications. While there is neuroscientific and biomedical literature that aims to address these challenges, we propose solutions that aim to enhance the functionality of BCIs as part of a larger system. We use several methods in order to tackle these fundamental roadblocks present in BCIs. By using transfer learning, we try to analyze ways for modifying an existing system trained on one user such that it performs well on unseen data. We also explore learning through limited labeled data using few-shot learning, as well as domain adaptation approaches to reshape target data so it appears similar to the source data in the probability space. We attempt to address low SNR in BCI signals by investigating how BCI signals respond to the presence of more than one stimulus and how these elicited signals interact with each other. We study how changes in one signal provide information about the other signal, which helps us with detection, attribution, synchronization, etc.