The open archive for STFC research publications

Full Record Details

DOI 10.5286/dltr.2021001
Persistent URL http://purl.org/net/epubs/work/48837027
Record Status Checked
Record Id 48837027
Title How surrogate models can enable integration of experiments, big data, modelling and simulation
Abstract AI surrogate models for science, based on Deep Neural Networks (DNN), are becoming increasingly popular. This is especially the case for generative models such as Autoencoders and Generative Adversarial Networks (GANs), which are showing promising results and exciting opportunities in several fields of science. We present here an overview of the current state of the art for surrogate modelling, starting with a broad view across several fields, ranging from particle methods for Molecular Dynamics to continuum solvers for Computational Fluid Dynamics. We then focus our attention to applications in Nuclear Fusion through plasma physics simulation, presenting the latest results and achievements. The paper not only outlines current limitations but also proposes the next steps in the research, in order to connect the high accuracy prediction power of the DNN together with scientific understanding and will detail the current state of their scalability on large computing resources. The main intention is to build a road map that bridges the traditional science and the novel big data in order to have solid and reliable surrogate models.
Organisation STFC , HC
Funding Information
Related Research Object(s):
Licence Information: Creative Commons Attribution 4.0 International (CC BY 4.0)
Language English (EN)
Type Details URI(s) Local file(s) Year
Report DL Technical Reports DL-TR-2021-001. STFC, 2021. DL-TR-2021-001.pdf 2021