Security of AI/ML
Investigating the unique vulnerabilities of AI and ML models, including adversarial attacks, data leakage, and defenses to ensure model integrity and privacy.
Focus Areas
-
Adversarial Attacks on AI Systems
Studying attacks against speaker verification, biometric systems, and other AI-driven security mechanisms using synthetic and adversarial inputs.
-
Model Privacy and Data Leakage
Investigating membership inference attacks and prompt-language leakage from LLM-generated data to understand information exposure risks in trained models.
-
Robustness of Biometric Models
Rethinking data assumptions and distribution models used in behavioral biometric systems to improve resilience against adversarial manipulation.
Related Publications
IEEE SVCC 2026
Prompt-Language Leakage from LLM-Generated Data via Membership Inference on Downstream LSTMs
IEEE SVCC 2025
SyntheticPop: Attacking Speaker Verification Systems With Synthetic VoicePops
Springer Nature 2025
Beyond Normality: Rethinking Behavioral Biometric Data