AID4GREENEST researchers have presented new research that strengthens the scientific foundations behind reliable, uncertainty-aware modelling, a key pillar for AI-driven innovation in sustainable industry.
The research: “A Framework for the Bayesian Calibration of Complex and Data-Scarce Models in Applied Sciences,” (currently available as a Preprint on arXiv) from Prof. Ignacio Romero and Dr. Christina Schenk at IMDEA Materials Institute, presents a comprehensive review and practical implementation of Bayesian calibration based on the well-established Kennedy-O’Hagan (KOH) framework.
This approach is widely recognised for its ability to combine simulation models and experimental data while rigorously quantifying uncertainty and accounting for model discrepancies.
For AID4GREENEST, this topic is especially relevant given the project’s focus on developing AI-supported modelling and rapid characterisation tools for greener steel production and manufacturing processes, where simulations are often computationally expensive and experimental data are limited.
Under these conditions, robust calibration and transparent uncertainty estimation are essential to ensure that predictive models can be trusted in research and industrial decision-making.
The publication also highlights the implementation and use of ACBICI (A Configurable BayesIan Calibration and Inference Package), an open-source Python library designed to translate the Bayesian calibration framework into a practical and extensible workflow.
The framework supports both single- and multi-output calibration, enabling simultaneous inference across several interrelated model outputs, a frequent requirement in complex materials and process simulations. It integrates advanced Bayesian inference algorithms, posterior approximation methods, Gaussian process–based prediction, and sensitivity analysis into a unified and largely automated pipeline.
This makes it possible to perform scalable, interpretable, and sample-efficient inference even in data-scarce scenarios.
A further contribution of the study is a consolidated set of practical recommendations for conducting reliable Bayesian calibration. These guidelines, derived from both theoretical insight and implementation experience, help users apply rigorous methods without unnecessary complexity.
This work from AID4GREENEST researchers Prof. Romero and Dr. Schenk lowers the barrier to adopting sophisticated calibration strategies. In doing so, it directly supports the project’s mission to combine AI, simulations, and experimental data in a scientifically robust way, accelerating the development of more sustainable materials and industrial technologies.