Secure Real-Time Systems
Timing leakage, schedule-based attacks, randomized scheduling, and real-time guarantees.
Research
Secure and intelligent cyber-physical systems under real-time constraints
My research focuses on securing real-time, embedded, and cyber-physical systems. I study timing-based information leakage, randomized scheduling and control stability, confidential AI execution, and learning-based vulnerability prioritization.
Cyber-Physical Systems Security
Security mechanisms that account for timing behavior, control stability, and safety constraints.
Trusted & Embedded AI
Confidential inference, resource-constrained ML, and learning-based vulnerability prioritization.
Selected research projects
Trusted execution · Real-time systems
Confidential DNN Execution inside Arm TrustZone
We develop scheduling and task-fusion techniques for confidential neural-network inference inside trusted execution environments while preserving real-time feasibility under tight secure-memory and context-switch constraints.
Secure inference pipeline
Normal World
→
TrustZone
+21.3%schedulable task sets at high utilization
~90%fewer TEE context switches
Timing leakage · Information flow
Timing-Based Information Leakage in Real-Time CPS
We investigate how observable scheduling behavior leaks information about higher-priority computation. The work spans multiframe timing channels and data-flow-driven cyber-physical systems, combining analytical models with statistical inference.
Observation model
High-priority task
⇢
Schedule traces
⇢
Inference
0.96example timing-pattern inference precision
2 papersmultiframe + data-flow leakage
Randomization · Control · Security
Security Analysis of Differential Privacy-Based Schedulers
We study how differential-privacy-inspired scheduling changes attack opportunities, deadline behavior, and closed-loop stability. Our ongoing work then asks how to retain useful schedule randomization while explicitly respecting control constraints.
Security–control tradeoff
Randomization
→
Attack surface
↔
Control stability
4attacks studied
Control-awarenext-stage scheduling design
Embedded AI · Performance
Language Models on Resource-Constrained Devices
We benchmark modern language models across embedded computing platforms and characterize the latency, memory, energy, and accuracy tradeoffs that determine whether a model is practical on-device.
Benchmarking dimensions
Model
→
Embedded platform
→
Profile
4+embedded platforms
4 metricslatency · memory · energy · accuracy
Cyber defense · Machine learning
Learning-Based Cyber Vulnerability Prioritization
We develop ranking and composite-risk methods that combine exploitability, severity, threat intelligence, and vulnerability text to help defenders identify high-risk CVEs earlier and allocate remediation effort more effectively.
Prioritization pipeline
SOTA vulnerability score
→
Learned ranking
→
Top CVEs
Multi-signalseverity + exploitability + text
Earlierprioritization before catalog inclusion
Interested in the details?
Browse the full publication list, follow the latest research outputs, or get in touch about collaboration.