Large-Neighbourhood Search for Optimisation in Answer-Set Solving

Authors: Thomas Eiter, Tobias Geibinger, Nelson Higuera Ruiz, Nysret Musliu, Johannes Oetsch, Daria Stepanova5616-5625

AAAI 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We experimentally demonstrate the effectiveness of clingo-lns on different benchmark problems.4 Unless stated otherwise, clingo was called with no additional command-line parameters, i.e, it uses a single solving thread and employs branch-and-bound-based optimisation.
Researcher Affiliation Collaboration 1 Institute for Logic and Computation, TU Wien, Favoritenstraße 9 11, 1040 Vienna, Austria, 2 CD-Lab Artis, TU Wien, Favoritenstraße 9 11, 1040 Vienna, Austria, 3 Bosch Center for AI, Robert Bosch Campus 1, 71272 Renningen, Germany
Pseudocode Yes Algorithm 1: LNS optimisation for a minimisation problem
Open Source Code Yes Our Python implementation of LNS with ASP in the loop, the solver clingo-lns, is publicly available.3 3http://www.kr.tuwien.ac.at/research/projects/bai/aaai22.zip.
Open Datasets Yes Weighted Strategic Companies... the instances are those of the 3rd (Calimeri, Ianni, and Ricca 2014), 4th (Alviano et al. 2013) and 5th (Calimeri et al. 2016) ASP Competition with random weights from [1, 1000] added.
Dataset Splits No The paper describes using
Hardware Specification Yes All experiments were run on a cluster with 13 nodes, each having 2 Intel Xeon CPUs E5-2650 v4 (max. 2.90GHz, 12 physical cores, no hyperthreading), with memory limit 20GB.
Software Dependencies Yes We used clingo v 5.5.1 and clingo-dl v 1.2.1.
Experiment Setup Yes The time limit to explore individual neighbourhoods was 20 seconds. The size of each neighbourhood was set to relax about 80% of the atoms over plays/3.