Staff profile

Milad Kazemi is an Assistant Professor in the Department of Computer Science at Durham University and a member of the JusTN0W programme, where he contributes to research on a just transition to net zero. His research operates at the intersection of counterfactual inference, reinforcement learning, and uncertainty quantification. Specifically, his work focuses on assume-guarantee reinforcement learning, conformal prediction, causal reasoning, and safety monitoring for Large Language Models (LLMs) and multi-agent systems. Ultimately, he aims to develop AI that is verifiable, uncertainty-aware, and safe for deployment in high-stakes, cyber-physical settings.

Prior to joining Durham University, Milad was a postdoctoral research fellow at King’s College London. He holds a PhD in Computer Science from Newcastle University, where his doctoral research focused on data-driven control synthesis for cyber-physical systems. His work is regularly published in leading venues, including AAAI, NeurIPS, JAIR, AAMAS, and USENIX.

• Formal Methods for Reinforcement Learning: Verification, synthesis, and symbolic reasoning to provide rigorous safety and performance guarantees for reinforcement learning systems.
• Trustworthy Reasoning with LLMs: Developing formal, symbolic, and neuro-symbolic methods to generate explanations, safety guardrails, and mathematically rigorous guarantees for large language models.
• Uncertainty Quantification: Leveraging distribution-free frameworks such as conformal prediction and "learn then test" to provide reliable, finite-sample probabilistic guarantees in machine learning.
• Counterfactual and Causal Inference: Applying algorithmic causal reasoning and counterfactual analysis to enhance the transparency, interpretability, and robust decision-making of autonomous systems.