Publications

A transformer-based decoder trained jointly across six intracortical speech BCI participants outperforms single-user models for every participant, cutting relative word error rate by over 50% on average. A new user can reach under 7% word error rate after finetuning on fewer than 200 sentences.

AdapSkin is a stretchable, bioadhesive conductive-polymer skin interface that adapts to age-related changes in skin such as dryness and wrinkles. By improving skin-electrode contact, it reduces variation in impedance, signal-to-noise ratio, and motion artifacts, yielding higher-quality surface EMG and better machine learning accuracy for neuromuscular assessment.

FlexEGG is a soft, stretchable abdominal electrode array built from a hybrid conductive-polymer and silver-nanowire conductor, enabling high-fidelity electrogastrography (EGG) monitoring of gastric myoelectrical activity in clinical and real-world settings.

A metadata-aware contrastive learning framework for predicting neurochemical concentrations from fast-scan cyclic voltammetry (FSCV). It models batch identity, scan sequence, and scan rate, and uses preference-based ranking losses to learn discriminative voltammetry features, improving accuracy over traditional baselines.

A high-frequency SSVEP-based brain-computer interface designed for use outside the lab. It combines dry-contact spring-loaded electrodes with active buffering, an actively driven guard, and a driven-right-leg circuit, plus an LED stimulus flickering above the threshold of visual perception for more comfortable everyday use.