Streamlining Nuclear Physics Data and Uncertainty Quantification with the Bayesian Mass Explorer

Obtaining up-to-date, reliable data is a necessary challenge across all domains of science, especially in nuclear physics. Nuclear data with uncertainties are often found in different places and formats, requiring significant effort to properly consolidate and compare. To address these challenges, the Bayesian Mass Explorer (BMEX) aims to provide an open-source suite of user-friendly web applications for on-the-fly data retrieval, visualization, and Bayesian uncertainty quantification. BMEX Masses, the project’s flagship app, focuses primarily on plotting experimental and model data, allowing for immediate model performance analysis and experimental feature extraction. BMEX also serves as a cloud-enabled stage for projects leveraging machine learning and advanced statistics, such as reduced basis methods and neural networks. In a current project, neural networks are trained to learn a normalizing flow that learns Bayesian posterior distributions of model parameters for a relativistic mean field mass model. These networks can then be deployed to a web app in BMEX to quickly sample parameters and generate quantified predictions. In the future BMEX will continue to provide new user-focused, accessible tools to the nuclear physics community in the fields of model emulation, online model calibration, and experimental design.