Reasoning about Connectivity Constraints

Authors: Christian Bessiere, Emmanuel Hebrard, George Katsirelos, Toby Walsh

IJCAI 2015 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We report results on several benchmark problems which demonstrate the efficiency of our propagation algorithms and the promise offered by reasoning globally about connectivity.
Researcher Affiliation Academia Christian Bessiere CNRS, U. Montpellier Montpellier, France bessiere@lirmm.fr Emmanuel Hebrard CNRS, U. Toulouse Toulouse, France hebrard@laas.fr George Katsirelos INRA Toulouse, France gkatsi@gmail.com Toby Walsh NICTA & UNSW Sydney, Australia toby.walsh@nicta.com.au
Pseudocode Yes Algorithm 1: DC on CONNECTIVITY (G, S)
Open Source Code Yes We implemented all techniques with minicsp 2http://www.inra.fr/mia/T/katsirelos/minicsp.html
Open Datasets No The paper describes generating instances for experiments (e.g., "We used the generator of Gomes et al. to create 20 random instances") but does not provide concrete access information (link, DOI, specific citation) for these generated instances or any other publicly available datasets used in a way that implies public access to the specific data splits.
Dataset Splits No The paper does not specify explicit train/validation/test dataset splits, sample counts for splits, or reference standard predefined splits for machine learning model training and validation.
Hardware Specification Yes We ran experiments on a cluster of 48-core Opteron 6176 nodes at 2.3 GHz with 378 GB RAM available.
Software Dependencies No The paper mentions "minicsp" as the software used for implementation but does not provide a specific version number for it.
Experiment Setup Yes We test a hybrid CP/SAT approach... We compare four models. The connectivity and cost minimization of the corridor is either decomposed with a CONNECTIVITY and a linear inequality (connect) or modeled directly with the weighted version (weighted). Then, in both cases, we try with or without face constraints (denoted +face ). We also report results without clause learning (denoted by ).