Description
In this presentation, I will introduce Vitess-AI, an initiative to enhance the VITESS neutron simulation application by integrating machine learning (ML) and large language model (LLM) technologies.
Vitess-AI aims to streamline two critical simulation workflows: (1) the optimization of complex, multi-parameter instrument configurations during the design phase, and (2) the preparation and planning of experiments on operational instruments through digital twin technologies.
By coupling VITESS with instrument control systems like NICOS and data sources such as NCrystal, the integration of LLMs will support data-driven optimization and exploratory scenario analysis.
In this first phase of the project, I will outline:
1. Key challenges related to domain knowledge representation, simulation validation, and model interpretability.
2. Points for discussion with the community, including the need for physics-informed ML, uncertainty quantification, and explainability tools to ensure transparency and trust in AI-assisted simulations.
3. An initial assessment of additional requirements for expanding the approach to other simulation platforms.
