Kyle Stein, Ph.D.
Artificial Intelligence × Cybersecurity

Kyle Stein, Ph.D.

Assistant Research Scientist · University of West Florida

I develop data-efficient, robust artificial intelligence for cybersecurity, with an emphasis on systems that generalize to unseen threats and continue adapting as attack behavior changes. My work spans transformer-based packet modeling, few-shot and continual learning, vision-language models, and backdoor defense.

I am open to collaborations with researchers at other universities on computer vision, the detection and mitigation of adversarial and backdoor attacks, continual and few-shot learning, and related AI × cybersecurity applications.

Explore Research Publications
Featured Research2026 · arXiv preprint

Temporal and Multimodal Deep Learning for Cyberattack Detection in LEO Satellite Systems

Kyle Stein, Guillermo Francia III, Eman El-Sheikh, Hossain Shahriar

Combining hardware, orbital, and radio-frequency telemetry to detect satellite cyberattacks across time.

Read paper PDF BibTeX
Hierarchical multimodal Transformer: hardware, orbital, and RF features are fused for each observation, then a temporal Transformer models the sequence before predicting the cyberattack class.
Figure 3. A subsystem Transformer combines the three telemetry sources; a temporal Transformer tracks how the satellite state evolves. Stein et al., 2026 · CC BY 4.0.
Research Focus

Building AI that adapts under uncertainty.

My research centers on robust learning under distribution shift, limited labels, emerging attack behavior, and changing operational environments.

01

AI for Cyber Defense

Transformer-based modeling of network traffic and cyber behavior for malware detection, threat recognition, and adaptive defense.

02

Continual & Few-Shot Learning

Parameter-efficient adaptation that learns new threats from limited examples while preserving previously acquired knowledge.

03

Trustworthy Vision-Language AI

Compositional reasoning, prompt-based adaptation, and open-world defenses for detecting previously unseen backdoor behavior.

Selected Work

Recent publications

View all publications →