معرفی
Meghdad Kurmanji is a Research Fellow at the University of Cambridge, working in the CaMLSys lab under Prof. Nic Lane. Their research focuses on machine unlearning, privacy, and trustworthy AI, with applications in computer vision, language models, and federated learning. They hold a PhD from the University of Warwick, where they developed novel solutions for updating machine learning models in learned database systems, addressing both continual learning and unlearning challenges.
Education: PhD in Computer Science at the University of Warwick (supervised by Prof. Peter Triantafillou). Prior work includes contributions to learned database systems and approximate query processing.
Research interests emphasize bridging gaps between theoretical machine unlearning and practical regulatory requirements, particularly in large language models (LLMs). Key themes include optimizing communication in foundation model training, designing scalable unlearning pipelines, and ensuring ethical AI compliance.
Recent publications highlight advancements in unlearning techniques for LLMs (e.g., LUNAR), communication-efficient distributed training (DES-LOC), and foundational analyses of unlearning challenges. Their work also explores dynamic data systems and out-of-distribution detection in learned databases.
Labs/Teams: Active member of the CaMLSys lab, collaborating on machine learning systems research.
