Machine Learning / Deep Learning

HAR using multiple sensors including RGB, depth, IMU, and event-based sensors with advanced feature fusion

Designing interpretable CNN-Transformer pipelines that estimate sleep stages from cross-modal biosignals collected in clinical studies and remote assignments.

A deep learning framework that detects mental stress directly from raw, ultra short-term ECG windows.

A wrist-actigraphy deep learning pipeline that beats classical sleep scoring without handcrafted features.

Applying CNNs and RNNs to ECG, PPG, and multi-sensor streams for real-time healthcare monitoring.

Document intelligence pipelines that detect figures, captions, and structured data using deep learning.