I develop robust methods for presentation attack detection (PAD) and liveness detection in contactless biometric systems, with emphasis on generalization to unseen attacks and domain shifts. My work includes GRU-AUNet, a domain-adaptation framework for contactless fingerprint presentation attack detection, as well as a universal anti-spoofing approach and an unsupervised deep learning method for contactless fingerprint biometric systems. I have also built competition-grade liveness detectors that ranked 2nd in the LivDet 2023 Noncontact Fingerprint and LivDet-Face 2024 challenges at IEEE IJCB. This line of work is further developed in my M.S. thesis, Deep Learning-Driven Biometric Security.

I apply deep generative models to camera-based vital sign estimation and cardiovascular feature analysis from video. In rECG, I proposed a guided diffusion framework (GEDM) to reconstruct ECG signals from facial video for contactless cardiac monitoring. In rPPG-SysDiAGAN, I proposed a GAN with a multi-domain discriminator to localize systolic and diastolic features in rPPG waveforms, improving interpretability and accuracy in contactless vital sign monitoring.

I integrate machine learning with physiological signals for continuous health monitoring and clinical applications. In Advancements in Continuous Glucose Monitoring: Integrating Deep Learning and ECG Signal, I study how ECG-based deep learning combined with CGM data can enable non-invasive, multimodal approaches to diabetes management and physiological monitoring.