Xiaolong Ma

Xiaolong Ma is currently a Postdoctoral Appointee in the Data Science and Learning (DSL) Division at Argonne National Laboratory, supervised by Dr. Rajkumar Kettimuthu. He received his Ph.D. in Computer Science from University of Nevada, Reno, advised by Prof. Feng Yan and Prof. Lei Yang. His research lies at the intersection of high-performance computing, distributed system and machine learning system, with a particular focus on building scalable, efficient, and reliable infrastructure for large-scale AI training and inference.

His recent work focuses on optimizing large language model inference on leadership-class computing systems, including Aurora and Polaris. His research explores adaptive and resource-efficient LLM serving, with topics such as prefill/decode disaggregation, dynamic batching and resource scaling, hierarchical KV-cache management, resilient and fault-tolerant inference, and the use of opportunistic or preemptible GPU resources. He is also interested in designing efficient and scalable multi-agent systems for scientific workflows, autonomous scientific discovery, and complex decision-making applications.

His broader research interests include distributed and parallel systems, cloud computing, storage systems, multi-agent systems, and AI for Science.