Security data is relational, adversarial, and voluminous, and its statistical properties change continuously over time as networks, software, and adversaries evolve. Together, these factors make cybersecurity an exceptionally challenging domain for machine learning.
My research focuses on machine learning that performs reliably in practice: models and agents grounded in domain expertise about networks and malicious software, rigorously evaluated, and interpretable by the analysts who rely on them.
Focus areas
LLMs and agentic systems
Large language models and agentic systems that reason over security data, use tools, and carry out multi-step analysis alongside human analysts.
Machine learning for security
Models that find, prioritize, and explain significant activity in large volumes of security data, from classical methods to deep learning.
Networks and malware
Learning the structure of traffic, protocols, binaries, and malicious code, and how threats evolve over time.
Rigorous evaluation
Feature engineering, benchmarking, and statistical discipline so models and agents hold up in practice.
Foundations
My machine learning work started with language: natural language processing, information extraction, and LDA topic modeling over large document collections, alongside ontologies and semantic knowledge graphs, including an ontology language that predated the W3C’s OWL. That background in language and structured knowledge carries directly into today’s work with large language models.
Publications and talks