Research Seminar by Dr Latifur Khan

This presentation explores the transformative potential of generative AI—particularly large language models (LLMs)—in addressing critical challenges in domains such as cybersecurity, intelligent transportation systems (ITS) and political sciences. In this talk, the speaker will cover the following and other related topics.

• Generative AI–Enhanced Threat Modeling in ITS: We develop an LLM-based framework to automate threat modeling for complex intelligent transportation systems by mapping information flows to MITRE ATT&CK techniques and NIST Cybersecurity Framework controls. The approach evaluates multiple AI methods, including zero-shot learning, RAG, multimodal reasoning, in-context learning, and fine-tuning.
• Policy Analysis for Secure Transportation Systems: This project enhances transportation cybersecurity policy using AI-driven legal analysis and stakeholder engagement. Building on the TraCR AI system, it integrates U.S. and international regulations and uses agentic AI and graph-based retrieval to identify policy gaps and propose improvements for data security and privacy in autonomous transportation.
• Conflict and Political Violence Monitoring: We develop a domain-specific pretrained language model for analyzing conflict and political violence data, which outperforms general-purpose LLMs in classification and question-answering tasks and has over 14,000 downloads on GitHub and Hugging Face. We also proposed ensemble-based active learning methods—Ensemble Union and Ensemble Intersection—that combine multiple heuristics to improve sample selection. Experiments on the United Nations Parallel Corpus show these approaches achieve performance comparable to full-dataset training while requiring far fewer labeled examples.