Circuit-Aware d-DNNF Compilation

Authors: Vincent Derkinderen, Jean-Marie Lagniez

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

Reproducibility Variable Result LLM Response
Research Type Experimental We empirically demonstrate the effectiveness of this approach, achieving a significant improvement in both the number of solved instances and the size of the resulting circuits. To empirically assess the benefits of eliminating don t care variables, we conducted experiments using 549 instances of probabilistic inference problems modeled as circuits.
Researcher Affiliation Academia 1KU Leuven, B-3000 Leuven, Belgium 2CRIL, U. Artois & CNRS, F-62300 Lens, France
Pseudocode Yes Algorithm 1: init Watches Input: a circuit Γ = N, E . Output: the set of gates that are not watched. 1 G {g N | (g , g) E} 2 Let watches be an empty map 3 for g N \ {g N | (g , g) E} do 4 watches[g] = {g N \ G | (g , g) E} 5 G G watches[g] 6 return N \ {G sink(Γ)}
Open Source Code Yes These instances, as well as the code, are available at [Derkinderen and Lagniez, 2025].
Open Datasets Yes To empirically assess the benefits of eliminating don t care variables, we conducted experiments using 549 instances of probabilistic inference problems modeled as circuits. These instances, as well as the code, are available at [Derkinderen and Lagniez, 2025].
Dataset Splits No The paper lists several datasets (dnf, noisy, bnkr, games, grid, raki, smokers, verilog) and states that experiments were conducted using 549 instances. However, it does not specify any training, testing, or validation splits for these datasets, nor does it describe a splitting methodology.
Hardware Specification Yes Our experiments ran on Intel Xeon E5-2643 processors at 3.30 GHz with 32 Gi B of RAM and Linux Cent OS.
Software Dependencies No The paper mentions evaluating the `d4` compiler and provides a GitHub link (https://github.com/crillab/d4v2), but it does not specify a version number for this compiler or any other key software dependencies.
Experiment Setup Yes Each instance had a time-out of 3600 seconds and a 32 Gi B memory limit.