Yahia H. Ali
BME PhD Defense Presentation

Date: 2026-02-13
Time: 10:30 AM-12:30 PM ET
Location / Meeting Link: HSRBII N600 / https://emory.zoom.us/j/98040009513?pwd=0VJmOBIl2u03VNDJSvccPUx3Nwcqn2.1

Committee Members:
Chethan Pandarinath, PhD (Advisor); Garrett Stanley, PhD; Michael Borich, DPT, PhD; Eva Dyer, PhD; David Brandman, MD, PhD


Title: Modular, low-latency brain-computer interfaces

Abstract:
Brain-computer interfaces (BCIs) are a promising avenue for restoring movement control to people with paralysis by decoding neural activity directly into actions. While researchers typically use linear models to map neural signals to cursor velocity, these decoders have limited capacity to learn neural population dynamics, constraining their ability to filter noise and generalize across datasets. Artificial neural networks (ANNs) can learn richer representations of neural state, but deploying them in real-time BCIs is challenging because decoding must occur on millisecond timescales to create an intuitive control experience. This dissertation advances BCI technology through three contributions spanning infrastructure, algorithms, and interaction paradigms. First, I present BRAND, a modular software platform for real-time BCI systems with full ANN support. BRAND achieves sub-millisecond inter-process communication and less than 8 milliseconds of end-to-end latency while supporting multiple programming languages, enabling rapid iteration on decoder designs in a modular system. BRAND has since been adopted by multiple BCI research groups working on clinical trials. Second, I deploy latent variable models in real-time BCI for the first time, finding that while these models match the performance of simple smoothing techniques, a fundamental labeling problem limits their ability to improve cursor decoding in a BCI setting. Third, I develop a discrete gesture-based navigation system that increases small target selection accuracy from less than 40% with cursor control to greater than 90%, establishing discrete control as a practical complement to traditional point-and-click interfaces. With these contributions, I aim to expand the design space for BCI control and provide the infrastructure necessary for rapid machine learning innovation, advancing BCIs toward practical, everyday assistive devices for people with paralysis.