[1S4-GS-2] Machine learning:

[1S4-GS-2-05] OpenBNSL: A Comprehensive Benchmarking Framework for Bayesian Network Structure Learning

〇Ryota Miyagi¹, Hirotaka Nishikori¹, Hiroshi Nakamura¹, Hideki Takase¹ (1. The University of Tokyo)

Keywords:Bayesian network, probabilistic graphical model, machine learning

A Bayesian network is a powerful probabilistic model that represents conditional independence relationships among variables using directed acyclic graphs. It has been widely used in various fields, such as medical diagnosis and fault detection. Learning the structure of a large Bayesian Network from data is computationally intensive, and optimization techniques, including parallelization, can significantly improve performance. In this study, we propose OpenBNSL, an open framework designed to enable fair and highly reproducible comparisons of Bayesian Network Structure Learning (BNSL) algorithms. OpenBNSL provides flexible implementations with a standardized evaluation process, a Docker-based reproducible experimental environment, and multi-core and many-core parallelization. Our framework ensures transparency by distributing the source code and evaluation procedures under the MIT License, demonstrating a strong commitment to the Open Science principle. In this paper, we establish an environment where various BNSL techniques can be systematically and fairly compared, thus supporting the further advancement of Bayesian Network research.