Perturbed Message Passing for Constraint Satisfaction Problems

Authors: Siamak Ravanbakhsh, Russell Greiner

JMLR 2015 | Venue PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Experimental results on random and real-world CSPs show that Perturbed BP is often more successful and at the same time tens to hundreds of times more efficient than standard BP guided decimation.
Researcher Affiliation Academia Department of Computing Science University of Alberta Edmonton, AB T6E 2E8, CA
Pseudocode Yes Algorithm 1: Belief Propagation-guided Decimation (BP-dec) Algorithm 2: Perturbed Belief Propagation
Open Source Code No The paper states: We implemented all the methods above for general factored CSP using the libdai code base (Mooij, 2010). This indicates the use of a third-party tool, not that the authors' specific implementation for the paper's methodology is open-sourced.
Open Datasets Yes We considered CSP instances from XCSP repository (Roussel and Lecoutre, 2009; Lecoutre, 2013)
Dataset Splits No The paper discusses using '100 random instances' for evaluation and benchmark CSP instances from the XCSP repository. However, it does not provide details on training/test/validation splits in the context of machine learning, as the algorithms are applied to solve individual problem instances rather than trained on a dataset to learn a model.
Hardware Specification No This research has been enabled by the use of computing resources provided by West Grid and Compute/Calcul Canada. This refers to general computing infrastructure but does not specify particular CPU or GPU models, memory, or detailed computer specifications.
Software Dependencies No We implemented all the methods above for general factored CSP using the libdai code base (Mooij, 2010). While it mentions 'libdai', it does not provide a specific version number for the software.
Experiment Setup Yes We used a convergence threshold of ϵ = .001 for BP and terminated if the threshold was not reached after T = 10 210 = 10, 240 iterations. To perform decimation, we sort the variables according to their bias and fix ρ fraction of the most biased variables in each iteration of decimation. This fraction, ρ, was initially set to 100%, and it was divided by 2 each time BP-dec failed on the same instance. ... For Perturbed BP, we set T = 10 at the starting attempt, which was increased by a factor of 2 in case of failure. ... For BP-dec and SP-dec, we use a convergence threshold of ϵ = .001 and fix ρ = 1% of variables per iteration of decimation. Perturbed BP and Perturbed SP use T = 1000 iterations.