CoCoA: A Non-Iterative Approach to a Local Search (A)DCOP Solver

Authors: Cornelis Jan van Leeuwen, Przemyslaw Pawelczak

AAAI 2017 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Experimentally, through evaluating graph coloring problems, randomized (A)DCOPs, and a sensor network communication problem, we show that Co Co A is able to very quickly find solutions of high quality with a smaller communication overhead than state-of-the-art DCOP solvers such as DSA, MGM-2, ACLS, MCS-MGM and Max-Sum.
Researcher Affiliation Collaboration Cornelis Jan van Leeuwen TNO and Delft University of Technology Eemsgolaan 3, Groningen, The Netherlands coen.vanleeuwen@tno.nl Przemysław Pawełczak Delft University of Technology Mekelweg 4, Delft, The Netherlands p.pawelczak@tudelft.nl
Pseudocode Yes The proposed algorithm is given in pseudocode in Algorithm 1, with accompanying set of messages and agent states in Table 1 and Table 2, respectively and discussed in detail in the subsequent section.
Open Source Code Yes For a replicability of results and figures, the source code is available upon request, or at https://github.com/coenvl/m SAM.
Open Datasets No The paper describes how problems/graphs are generated for each experiment (e.g., 'The graphs are generated by selecting n = 500 random points in two dimensional space using a Poisson point process'), rather than using or providing access to a pre-existing public dataset.
Dataset Splits No The paper describes how experimental problems are generated and evaluated, but does not specify explicit training, validation, or test dataset splits.
Hardware Specification Yes The experiments are carried out on a laptop with an Intel Core i7-3720 CPU 2.6 GHz and 8 GB RAM.
Software Dependencies Yes The solvers are implementated in Java 1.7, and the experiments are set up in Matlab 2015b, which is also used to post-process and present the result figures.
Experiment Setup Yes DSA (variant C, with update probability, p = 0.5), MGM-2 (Maheswaran, Pearce, and Tambe 2004) (with offer probability p = 0.5), Max-Sum ADVP (Zivan, Parash, and Naveh 2015) from hereon also referred to as simply Max-Sum, (switching graph direction after 100 iterations, value propagation after two switches, and using the constraint standard inner order), ACLS (with update probability p = 0.5) and MCS-MGM (Grubshtein et al. 2010) (non-parametric).