LLM-assisted explainability for neural combinatorial optimization and VAE-based causal generative models for intracranial EEG.

Developed an end-to-end simulation and optimization platform integrating physical modeling with ML-driven inverse optimization under biophysical constraints, achieving >100x efficiency over SOTA methods.

Designing ML pipelines for learning predictive biomarker representations from high-dimensional neural recordings, source localization via convex optimization and deep learning, and posterior inference using conditional diffusion models.