Projects

SODEEP: A Lightweight Causal CNN for Real-Time Slow Oscillation Detection

Causal CNN (~100k params) achieves F1 of 0.967 with 39 ms latency, replacing classical methods for closed-loop sleep stimulation.

PyTorch Causal CNN LSTM Real-Time

Riemannian Geometry Boosts fNIRS Brain-State Classification

Block diagonal SPD covariance kernels for fNIRS. 89% two-class and 70% eight-class accuracy, outperforming traditional approaches.

SPD Manifolds pyRiemann fNIRS Block Diag. Kernels

fNIRS-Based Awareness Detection in Disorders of Consciousness

Individualized Riemannian pipeline for DoC patients. 100% sensitivity and 89% specificity for awareness detection at bedside.

Clinical BCI Riemannian fNIRS MNE

Riemannian Manifold Analysis of Infant Memory Formation

Three-month-olds formed new voice memories after familiarization and sleep. Riemannian K-means clustering on SPD covariance matrices revealed convergence of voice representations.

pyRiemann SPD Manifolds Laplacian Eigenmaps LME Models

GazeAnnot: Interactive Eye Movement Labeling & U'n'Eye Training

PyQt6 desktop app for labeling saccades, microsaccades, and PSO in eye tracking data, with integrated U'n'Eye deep learning model training and evaluation.

PyQt6 PyTorch Eye Tracking Deep Learning