Local Bayesian Dirichlet mixing of imperfect models

Journal Article · 10 November 2023 · Scientific Reports

DOI: https://doi.org/10.1038/s41598-023-46568-0 · OSTI ID:2205360

Kejzlar, Vojtech  ^[1]; Neufcourt, Léo  ^[2]; Nazarewicz, Witold  ^[2]

---

1. Skidmore College, Saratoga Springs, NY (United States)
2. Michigan State Univ., East Lansing, MI (United States)

To improve the predictability of complex computational models in the experimentally-unknown domains, we propose a Bayesian statistical machine learning framework utilizing the Dirichlet distribution that combines results of several imperfect models. This framework can be viewed as an extension of Bayesian stacking. To illustrate the method, we study the ability of Bayesian model averaging and mixing techniques to mine nuclear masses. We show that the global and local mixtures of models reach excellent performance on both prediction accuracy and uncertainty quantification and are preferable to classical Bayesian model averaging. Additionally, our statistical analysis indicates that improving model predictions through mixing rather than mixing of corrected models leads to more robust extrapolations.

• Research Organization: Michigan State Univ., East Lansing, MI (United States)
• Sponsoring Organization: USDOE Office of Science (SC), Nuclear Physics (NP); National Science Foundation (NSF); USDOE
• Grant/Contract Number: SC0023688; SC0013365; 2004601
• OSTI ID: 2205360
• Alternate ID(s): OSTI ID: 2217277
• Journal Information: Scientific Reports, Vol. 13, Issue 1; ISSN 2045-2322
• Publisher: Nature Publishing GroupCopyright Statement
• Country of Publication: United States
• Language: English