Embedded Files
My research focuses on computer vision and trustworthy AI for high-stakes decision making, with an emphasis on identity protection. With 15 years of experience, I address challenges in AI-driven security, including bias, adversarial attacks, explainability, and privacy. My work centers on advanced algorithms for complex multivariate data such as degraded signals, hyperspectral imagery, biometrics, and video, mitigating threats including sensor spoofing and identity fraud. I have contributed to pioneering liveness detection in fingerprint recognition and to the use of hyperspectral imaging and sweat-based physiological signals in identity sciences.
Machine Unlearning
Machine Unlearning
This research develops efficient machine unlearning methods to selectively remove sensitive data, such as identity traits or biased text. It advances fairness and compliance in biometric and language models.
Security in Virtual Reality
Security in Virtual Reality
This research enhances identity verification in Mixed Reality by utilizing iris recognition with synthetic data for spoof resistance and multimodal fusion for improved accuracy.
Hyperspectral Imaging
Hyperspectral Imaging
This technology combines imaging and spectroscopy to capture rich spectral data across hundreds of wavelengths, enabling deeper analysis for advanced recognition, detection, and decision-making in defense contexts.
Inclusive Cybersecurity
Inclusive Cybersecurity
This project explores optics-informed AI to develop a contactless biometric security system that reduces skin tone bias through explainable models, enhancing mobile security.
Mobile Security
Mobile Security
This project enhances the detection of presentation attacks in finger photo recognition technologies, combining various color space representations.
Explainable AI (XAI)
Explainable AI (XAI)
This research examines XAI to promote transparency, enhance control over AI systems, and foster public trust by clarifying the decision-making process.
Chemical Biometrics
Chemical Biometrics
This exploratory study investigates human sweat as a non-invasive, spoof-resistant biometric by developing and evaluating protocols that address factors impacting biomarker reliability such as sampling site, timing, and individual variability.
Usable Biometric MFA
Usable Biometric MFA
The project focuses on developing robust multi-factor authentication using biometric factors; I developed FingerPIN, which combines sequential finger scans with PIN entry to enhance security —pilot results show strong usability.
Video Analysis
Video Analysis
Video data offers richer information than single images, enhancing accuracy and attack resilience. This research develops deep learning algorithms for robust identity matching from short video sequences.
Capture Bias Mitigation
Capture Bias Mitigation
Device diversity has a significant impact on the accuracy of security algorithms. Addressing this variability is essential for reliable and trustworthy decision-making systems.
Forensics
Forensics
Analysis of FBI fingerprint data shows that match performance varies with demographic covariates such as age and gender. Incorporating these factors via ROC regression provides a more accurate measure of discriminatory capacity.
DNA Profiling
DNA Profiling
DNA identification relies on 13 STR alleles, but detecting them in degraded samples is challenging with conventional or Rapid DNA systems. Our adaptive signal-processing algorithm effectively detects allelic peaks in degraded DNA.
Google Sites
Report abuse