Accelerator Seminar - Antonin Sulc

On Thursday, March 13, at 11 a.m., Antonin Sulc of Lawrence Berkeley National Laboratory will present "Addressing Challenges in Accelerator Operations With Artificial Intelligence" via Zoom.

Abstract: The evolution of particle accelerators into increasingly complex systems presents challenges and opportunities for artificial intelligence integration. In this talk, I will present our ongoing work towards a paradigm shift in accelerator operations through natural language processing and AI. At the core of our research is developing a decentralized multi-agent framework powered by large language models (LLMs), where specialized agents collaborate to control individual accelerator components to achieve increasingly autonomous operation. This framework is complemented by our efforts to unlock valuable operational insights from facility logbooks using retrieval augmented generation (RAG), which aims to better understand the underlying problems, look for root causes, and perform prevention from the information that we gathered over decades of operations. In the second part of the talk, I will discuss our advances in anomaly detection, demonstrating how our neural network approach enables robust fault detection even with limited labeled data -- a common challenge in accelerator operations. The talk will conclude with insights from our work on fine-tuning LLMs for accelerator-specific applications and outline future directions for integrating these technologies into accelerator facilities.