Maximum A Posteriori Inference in Sum-Product Networks

Authors: Jun Mei, Yong Jiang, Kewei Tu

AAAI 2018 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Our comprehensive experiments on real-world datasets show that our exact solver runs reasonably fast and could handle SPNs with up to 1k variables and 150k arcs within ten minutes; our approximate solver provides a good trade-off between speed and accuracy and has better overall performance than previous approximate methods.
Researcher Affiliation Academia Jun Mei, Yong Jiang, Kewei Tu Shanghai Tech University {meijun,jiangyong,tukw}@shanghaitech.edu.cn
Pseudocode Yes Algorithm 1 Calculate MAP2MAXS(Q, e, H)
Open Source Code Yes Our code is available at https://github.com/ shtechair/maxspn.
Open Datasets Yes We evaluated the MAP solvers on twenty widely-used real-world datasets (collected from applications and data sources such as click-through logs, plant habitats, collaborative filtering, etc.) from (Gens and Domingos 2013)
Dataset Splits No The paper describes how variables are divided into query, evidence, and hidden variables for generating MAP problems (e.g., 'Q/E/H proportion being 3/3/4'), but it does not specify explicit train/validation/test dataset splits for the learning process or overall experimentation.
Hardware Specification Yes We ran our experiments on Intel(R) Xeon(R) CPU E5-2697 v4 @ 2.30GHz.
Software Dependencies No The paper does not provide specific version numbers for software components or libraries used in the experiments.
Experiment Setup Yes When running the solvers, we bounded the running time for one MAP problem by 10 minutes. For BS, we tested beam sizes of 1, 10 and 100. For KBT, we tested K = 10 and 100.