Projects & Areas

A security system for distributed environments where transaction integrity is guaranteed across both parties. No single participant — regardless of their access level — can silently manipulate or fabricate a valid record. The system is designed to make fraud structurally impossible, not just detectable.

• Prevents unilateral transaction forgery — neither client nor server can fake a valid record alone
• Any tampering on either side of the ledger is automatically detectable by the other
• Resistant to replay attacks, identity spoofing, rollback to previous states, and concurrent fraudulent submissions
• Provides mutual auditability: both parties can independently verify the integrity of the full transaction history

Unpublished. Paper under preparation — methodology details withheld.

Network Intrusion Detection Systems degrade over time as attackers evolve and network environments shift. This research addresses performance degradation in ML-based NIDS by combining Model-Agnostic Meta-Learning (MAML) with Open-Set Recognition (OSR) — enabling models to rapidly adapt to new attack patterns with minimal retraining data while recognizing and safely handling unseen threats.

• Open-Set Recognition to flag and quarantine unknown attack patterns rather than misclassify them
• MAML for few-shot model repair: recover performance on degraded tasks without full retraining
• Empirical analysis of continual learning strategies (regularization, replay, MAML) against catastrophic forgetting in network security contexts
• Evaluated on real-world network intrusion datasets across multiple attack families

An undergraduate research survey reviewing a decade of ECG analysis studies, with a focused examination of the preprocessing pipelines used before classification, anomaly detection, and arrhythmia analysis. Preprocessing is shown to be a critical and often underspecified step that significantly impacts model performance.

• Reviewed neural network and machine learning approaches to ECG-based cardiac analysis
• Systematically compared preprocessing strategies (filtering, segmentation, normalization) across studies
• Published at SIU 2021 — the IEEE Signal Processing and Communications Applications Conference

Microservice containerization introduces unique security challenges due to weaker isolation compared to virtual machines. This work empirically evaluates host-based (syscall monitoring) versus network-based IDS strategies for containerized deployments, providing a principled comparison to guide secure microservice architecture decisions.

• Benchmarked multiple IDS sensor types against containerized workloads
• Analyzed trade-offs between host-based syscall monitoring and network-level traffic analysis
• Published at the IEEE Workshop on Secure and Reliable Microservices and Containers (SRMC 2022)