AI-Driven Security Research
I am fascinated by the disciplines of Artificial Intelligence, Machine Learning, Pattern Recognition, and Image Processing.
I am passionate about Security and Biometrics, the Science of Who You Are.
Trustworthy Identity Verification - Biometric technologies have been increasingly adopted for identity verification in several critical applications. However, several outstanding concerns create mistrust in their use. Trust depends on various factors, including system performance, security, fairness, explainability, and privacy. The impact of demographic differentials and vulnerability to presentation attacks directs the research toward finding unbiased and more secure representations, exploring new modalities, and different types of sensors to acquire the biometric raw signal.
Inclusive Cybersecurity
Technological discrimination, or the inability of AI-powered security systems relying on optical sensors to properly extract salient features, is a challenge marginalized people face every day. This project aims to develop a contactless biometric mobile security application that can mitigate the vulnerabilities of deep AI and optical sensors and allow marginalized identities the same access to data security.
Hyperspectral Biometrics
Hyperspectral imaging (HSI) can lead to broad, innovative work to establish human identity by combining imaging with spectroscopy. HSI generates data in hundreds of wavelengths, providing richer features to create a deeper identity profile and, subsequently, more accurate and resilient systems to spoofing.
Mobile Biometrics
Although smartphones have developed into highly portable and accelerated computing devices, securing them properly is a growing challenge. Finger photo recognition represents a promising touchless technology that offers portable and hygienic smartphone authentication solutions, eliminating physical contact.
Explainable AI
Explainable AI (XAI) provides an opportunity to comprehend the results generated by AI systems. Justifying AI decisions increases AI control and represents an opportunity to improve AI algorithms. Explaining these algorithms can contribute to deepening society’s trust in biometric technology and understanding causal relationships to obtain helpful information from the model that can be successfully used in other applications.
Video Analysis
Confirming an individual's identity by processing a single biometric image captured in a mobile environment is arduous. The acquisition's unconstrained nature is characterized by challenges such as background, pose, illumination variations, and evolving presentation attacks (PAs). These issues can be mitigated by exploiting video data that contain more information for the system to process and more robust to attacks.
Mitigating Capture Bias
Creating trustworthy decision-making systems is essential in critical applications impacting human beings. Capture bias is related to how the images are acquired, both in terms of the device used and the collector's preferences for point of view, lighting conditions, etc. Changes in these characteristics contribute to PAD performance degradation. If addressed, this concern can increase trust.
Usable AI Security Systems
Usability is not a one-dimensional property of a system; it is a combination of factors, including how quickly an experienced user can complete tasks, whether a user can remember how to use the system effectively in future visits, how frequently users make errors while using the system, how users recover from the mistakes, and whether the user enjoys using the system.