Maham Fatima

Developing a neuro-symbolic reinforcement learning framework for adaptive medical triage in mass casualty incidents. Integrating symbolic triage rules, multi-agent RL, and explainable AI for real-time decision-making.

Designed hyperspectral imaging acquisition pipeline for PPG-based heart rate estimation. Adapted PhysNet deep learning model to multi-band inputs, reducing error from 16 bpm (RGB) to 5 bpm.

Designed a synchronized acquisition pipeline for facial multispectral imaging and ground-truth PPG signals; enabling dataset creation, timestamp alignment, and validation. Adapted and optimized the PhysNet deep learning model for 3-channel RGB and multi-band hyperspectral inputs, including architectural modifications and experimental tuning across multiple datasets. Conducted experiments comparing RGB and multispectral imaging, involving preprocessing, dimensionality reduction, augmentation, and signal post-processing. Led spectral band selection and generalization analysis, showing that optimized bands and augmentation improve accuracy while supporting remote health monitoring and fairness across skin tones.