Bayesianism in the Age of Modern AI: ALC Present Uncertainty-Aware Surrogate Modelling and Optimisation Research
10 Nov 2026
Xingchi Liu:
I am the Acting Group Leader of the AI Smart Facilities team within STFC’s Scientific Computing Department. My research focuses on machine learning-based surrogate modelling, data-driven black box optimisation and uncertainty quantification, particularly how prior knowledge can be used to improve data-efficient optimisation, support decision-making and enable rapid model adaptation.
I am currently working with facilities including Diamond Light Source and the Central Laser Facility (CLF), developing surrogate models of complex scientific instruments and exploring optimisation approaches to improve their performance and operational efficiency.
I attended the BayesAI workshop with my colleague Dr. Tamilarasan Sabapathy – Rethinking the Role of Bayesianism in the Age of Modern AI, held at the University of Edinburgh from 26–30 October 2026.
The workshop brought together researchers to explore how Bayesian machine learning can improve the reliability and trustworthiness of modern AI. Topics included uncertainty quantification, incorporating prior knowledge, robust decision-making, model adaptation and applications of Bayesian methods to large language models (LLMs).
Several topics were closely aligned with the challenges we face at STFC, particularly uncertainty-guided optimisation, data-efficient model adaptation and the use of generative models to identify promising regions in high-dimensional optimisation problems. These approaches could help improve the efficiency and reliability of scientific instrument optimisation and support ongoing projects involving surrogate modelling and virtual accelerators.
Another interesting theme was the use of Bayesian methods to quantify uncertainty in LLMs and improve the reliability of their predictions and reasoning. This could also be relevant to the wider development of trustworthy AI agents for scientific research and discovery.
Overall, the workshop provided valuable insights into emerging Bayesian machine learning approaches and potential opportunities to apply them to real-world challenges across STFC’s scientific facilities.
Poster Presentation
Xingchi Liu – Towards Smart Scientific Facilities: Uncertainty-Aware Modelling and Optimisation
I also presented a poster highlighting our research on uncertainty-aware surrogate modelling and optimisation, and how these methods can contribute to more efficient and reliable operation of large-scale scientific facilities.
