New Compilation Languages Based on Restricted Weak Decomposability

Authors: Petr Illner

AAAI 2025 | Venue PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Our experiments demonstrate that nw DNNF circuits are suitable for computing MPEs in two-layer Bayesian networks (BNs) with large domains.
Researcher Affiliation Academia Petr Illner Department of Theoretical Computer Science and Mathematical Logic, Faculty of Mathematics and Physics Charles University, Czech Republic EMAIL
Pseudocode No We refer the curious reader to (Illner and Kuˇcera 2024) for a more detailed description of Bella and the pseudocode, where the methods compute Components and compute New Cut were changed as described below.
Open Source Code Yes 1https://github.com/Illner/Bella Compiler
Open Datasets No We implemented Bels2 to (randomly) generate and convert such BNs into CNF formulae using the encodings mentioned in this paper.
Dataset Splits No Ten instances were created for each density. Since the compilers are randomised, each instance was compiled three times, and the given results are averages.
Hardware Specification Yes The experiments3 were performed on a Linux machine (Debian 11) using an AMD EPYC 7543 2.8GHz processor and 512 Gi B of RAM.
Software Dependencies No The experiments3 were performed on a Linux machine (Debian 11) using an AMD EPYC 7543 2.8GHz processor and 512 Gi B of RAM. The time-out (resp. memory-out) was set to two/six hours (resp. 16 GB).
Experiment Setup Yes The time-out (resp. memory-out) was set to two/six hours (resp. 16 GB). The following compilers were considered: Bella, D44 (the randomised variant introduced by Illner and Kuˇcera (2024) was used), C2D5, and Sharp SAT-TD6. Ten instances were created for each density. Since the compilers are randomised, each instance was compiled three times, and the given results are averages.