Robust BCIsRobust BCIs with Uncertainty Estimates using Bayesian Neural Networks

Brain-Computer Interfaces (BCIs) offer users the ability to control devices directly from brain signals, thus bypassing the peripheral nervous system. Those brain signals are noisy and non-stationary and require robust machine learning systems to achieve good control. This project explores the ability of Bayesian Neural Networks to indicate when the decoding is unreliable and a predicted control command should not be executed. Such Bayesian methods offer unique opportunities to identify variabilities in a deployed BCI that would otherwise be neglected. This will allow us to achieve more robust control in more diverse use contexts.