BCI Lab

Non‑invasive brain–computer interfaces (BCIs) use external sensors to read brain activity without surgery. Non-invasive techniques rely on EEG, fNIRS, MEG, or other surface‑level neuroimaging to decode intention.  Recent research shows major improvements in signal quality, deep‑learning decoding, and real‑time control of robotic and digital systems.

The BCI Lab advances studies in the following areas:

• Robotic control: Non‑invasive BCIs to reliably control cursor movements, robotic arms, wheelchairs, and drones for executing complex tasks.
• Deep learning & manifold decoding: Modern BCIs using Convolutional Neural Networks (CNNs), Recurrent Neural Networks (RNNs), transformers, and manifold‑learning approaches to decode motor imagery and cognitive states with higher accuracy. 
• Transfer learning: Reducing calibration time and enabling “plug‑and‑play” BCI integration.
• Open‑source ecosystems: Contribute to toolboxes such as  OpenBCI, BCI2000, and MNE‑Python to accelerate experimentation and accessibility. 
• Holistic frameworks: integrating multimodal signals, adaptive algorithms, and user‑state monitoring to improve reliability and ease of use. 

Computer‑to‑Brain Interfaces (CBI) are systems that deliver information or stimulation into the brain, rather than reading signals out of it. They are the “reverse direction” of BCIs and represent the frontier of neuromodulation, sensory restoration, and closed‑loop therapeutic systems.  BCI Lab is working to evolve CBIs from crude stimulation into precise, targeted, adaptive neural‑modulation technologies using ultrasound, magnetic, electrical, optical, and implantable approaches.

Examples of these approaches are: 

• TMS (transcranial magnetic stimulation)
• Ultrasound neuromodulation
• Intracortical microstimulation (ICMS)

Sub-vocal Electromyography (EMG) is the core technology behind modern silent‑speech interfaces: systems that decode speech‑related muscle activity even when you don’t make audible sound.  Surface EMG on the face/neck can now be translated directly into text or commands using deep‑learning models, without needing any audio. This makes it ideal for hands‑free control, privacy‑sensitive communication, and hybrid BCI systems. 

Sub‑vocal EMG captures tiny electrical signals from articulatory muscles—lips, jaw, tongue base, larynx—when you silently articulate words. These signals occur even when no sound is produced.