Research Projects
My work connects real-time systems, cyber-physical systems security, and trustworthy AI. My work spans confidential deep-learning inference in trusted execution environments, information leakage in fixed-priority systems, randomized scheduling and control stability, embedded language-model performance, and learning-based cyber vulnerability prioritization.
Trusted execution · Real-time systems
We developed scheduling and analysis techniques for running confidential neural-network inference inside Arm TrustZone while preserving real-time guarantees. The work considers secure-memory limits and reduces costly trusted-execution context switches.
Read the ECRTS paper →Covert channels · Information leakage
We investigate how a low-priority task can infer the execution pattern of a higher-priority task from timing observations. Our work develops analytical models and statistical techniques for detecting covert channels in fixed-priority systems.
Read the ISORC paper →
Privacy · Scheduling · Control
We study how randomized scheduling affects security, deadline guarantees, and control stability. The analysis shows that reducing predictability can also create new anterior, posterior, and pincer attack opportunities, so privacy must be evaluated together with system-level safety.
Read the EMSOFT paper →Embedded AI · Performance
We benchmark language models on Raspberry Pi, Jetson, UP², and UDOO platforms. The study compares inference latency, memory use, energy consumption, and accuracy to help designers select practical model and hardware combinations.
Read the ICPE paper →
Cyber defense · Machine learning
We develop composite risk metrics that combine severity, exploitability, weakness information, threat intelligence, and semantic signals. The goal is to help defenders identify high-risk CVEs earlier and allocate remediation resources more effectively.
Ongoing research