Machine Learning for Neural Sensing

Overview

Research conducted at Carnegie Mellon University on machine learning and statistical modeling for neural sensing systems.

Key contributions:

• Designed an end-to-end machine learning pipeline to learn predictive biomarker representations from high-dimensional neural recordings.
• Analyzed and quantified theoretical limits of EEG sensing systems with Fisher information metrics.
• Built a convex-optimization-based seizure source localization framework using graph Laplacians and spectral clustering, achieving ~1 cm localization error on clinical data.
• Applied and tuned a CNN-LSTM spatiotemporal deep learning framework for EEG-based seizure imaging, achieving 3x lower localization error than SOTA optimization methods.
• Developed a conditional diffusion model and optimization framework for posterior inference in high-dimensional inverse problems, validating uncertainty recovery using analytically tractable Gaussian benchmarks.