Transformer-Based Cybersecurity
Self-supervised and supervised transformer models for packet inspection, network threat recognition, and malware detection, including settings where only a few labeled examples are available.
Research
Adaptive AI for evolving cyber threats.
I study how artificial intelligence systems can learn from limited data, remain robust under distribution shift, and adapt to threats that were not represented during initial training.
Few-Shot & Continual Learning
Learning new classes and emerging behaviors from small amounts of data while mitigating catastrophic forgetting and limiting computational overhead.
Parameter-Efficient Adaptation
Adapters, prompts, and lightweight model updates that specialize large pretrained models without requiring full-model retraining.
Backdoor & Adversarial Defense
Open-world defenses that use vision-language models and compositional reasoning to identify novel backdoor triggers and previously unseen attack patterns.
Compositional Learning
Methods that reason over reusable primitives such as objects and states to improve generalization when new combinations appear at test time.
Critical Infrastructure Security
Applied AI research for cyber-physical systems and critical infrastructure, informed by work on hydroelectric-dam testbeds, intrusion detection, and operational cyber defense.
Dissertation
Integrating Transformers for Cyber Defense Under Unseen and Evolving Threats with Deep Packet Inspection, Zero-Shot Backdoor Detection, and Continual Learning
Access dissertationThe dissertation brings together transformer-based packet inspection, zero-shot reasoning for unseen backdoor behavior, and continual learning for evolving threat environments.