In the absence of benchmark datasets for inference algorithms in probabilistic relational models, we propose an extendable benchmarking suite named ComPI that contains modules for automatic model generation, model translation, and inference benchmarking. The functionality of ComPI is demonstrated in a case study investigating both average runtimes and accuracy for multiple openly available algorithm implementations. Relatively frequent execution failures along with issues regarding, e.g., numerical representations of probabilities, show the need for more robust and efficient implementations for real-world applications.
CITATION STYLE
Potten, T., & Braun, T. (2020). Benchmarking Inference Algorithms for Probabilistic Relational Models. In Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) (Vol. 12277 LNAI, pp. 195–203). Springer Science and Business Media Deutschland GmbH. https://doi.org/10.1007/978-3-030-57855-8_15
Mendeley helps you to discover research relevant for your work.