Approximate Knowledge Compilation by Online Collapsed Importance Sampling

Authors: Tal Friedman, Guy Van den Broeck

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

Reproducibility Variable Result LLM Response
Research Type Experimental Our experimental evaluation shows that collapsed compilation performs well on standard benchmarks.
Researcher Affiliation Academia Tal Friedman Computer Science Department University of California Los Angeles, CA 90095 tal@cs.ucla.edu Guy Van den Broeck Computer Science Department University of California Los Angeles, CA 90095 guyvdb@cs.ucla.edu
Pseudocode Yes Algorithm 1: Online Collapsed IS
Open Source Code Yes We provide an open-source Scala implementation of this collapsed compilation algorithm.1 The code is available at https://github.com/UCLA-Star AI/Collapsed-Compilation. It uses the SDD library for knowledge compilation (Darwiche, 2011) and the Scala interface by Bekker et al. (2015).
Open Datasets Yes From the 2014 UAI inference competition, we evaluate on linkage(1077,1077), Grids(100,300), DBN(40, 440), and Segmentation(228,845) problem instances. From the 2008 UAI inference competition, we use two semi-deterministic grid instances, 50-20(400, 400) and 75-26(676, 676).
Dataset Splits No No specific dataset split information for training or validation was provided.
Hardware Specification No No specific hardware details (e.g., GPU/CPU models, memory amounts, or detailed computer specifications) were provided for running experiments.
Software Dependencies No The code is available at https://github.com/UCLA-Star AI/Collapsed-Compilation. It uses the SDD library for knowledge compilation (Darwiche, 2011) and the Scala interface by Bekker et al. (2015).
Experiment Setup Yes For evaluation, we run all sampling-based methods 5 times for 1 hour each. We report the median Hellinger distance across all runs... and ...we compare the performance on three different settings for the circuit size threshold: 10,000, 100,000, and 1,000,000. To constrain EDBP, we limit the corresponding circuit size for the junction tree used. In our experiments we set these limits at 100,000 and 1,000,000. To constrain SS, we limit treewidth w at either 15, 12, or 10.