Distributed Bayesian: A Continuous Distributed Constraint Optimization Problem Solver
Authors: Jeroen Fransman, Joris Sijs, Henry Dol, Erik Theunissen, Bart De Schutter
JAIR 2023 | Venue PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | The performance of the algorithm is evaluated empirically based on the sample efficiency. The proposed algorithm is compared to state-of-the-art DCOP and C-DCOP solvers. The algorithm generates better solutions while requiring fewer samples. Lastly, simulation results are given for randomly generated graphs and sensor coordination problems to compare the sample efficiency of D-Bay to state-of-the-art DCOP and C-DCOP solvers. |
| Researcher Affiliation | Collaboration | Jeroen Fransman EMAIL Delft Center for Systems and Control (DCSC), Delft University of Technology Joris Sijs EMAIL Delft Center for Systems and Control (DCSC), Delft University of Technology Henry Dol EMAIL Netherlands Organisation for Applied Scientific Research (TNO) Erik Theunissen EMAIL Netherlands Defence Academy (NLDA) Bart De Schutter EMAIL Delft Center for Systems and Control (DCSC), Delft University of Technology |
| Pseudocode | Yes | D-Bay as described in Algorithm 2 (Appendix A) involves four phases: ... Algorithm 2: Distributed Bayesian (D-Bay) for agent ai |
| Open Source Code | Yes | Implementations of the proposed algorithm and the state-of-the-art DCOP and C-DCOP solver have been added to the open-source software library pyDCOP (Rust et al., 2019) and made available publicly2. 2. https://gitlab.com/jfransman/pyDcop/ |
| Open Datasets | No | For the generation of the random graph experiments, the NetworkX (Hagberg, Schult, & Swart, 2020) generator, embedded within the pyDCOP library, is used. ... The locations of the targets t are uniformly distributed within the combined observation domains of the sensors. |
| Dataset Splits | No | For the generation of the random graph experiments, the NetworkX (Hagberg, Schult, & Swart, 2020) generator, embedded within the pyDCOP library, is used. Based on the randomly created graph, a C-DCOP is generated by allocating a variable (and agent) to every node and defining utility functions for all edges. ... All experiments are repeated 50 times and the median of the most illustrious results are shown in Figure 8. ... The results for 30 randomly generated problems for 6 sensors and 12 targets. |
| Hardware Specification | Yes | The simulations are conducted on a 2.1 GHz Intel Xeon Gold 6152 CPU machine with sufficient memory for the requirements of all solvers and the computation time is limited to one hour. |
| Software Dependencies | No | The implementation of DPOP is included within the pyDCOP library (Rust, Picard, & Ramparany, 2019). All other algorithms have been included in the pyDCOP library and made available publicly1. ... For the generation of the random graph experiments, the NetworkX (Hagberg, Schult, & Swart, 2020) generator, embedded within the pyDCOP library, is used. |
| Experiment Setup | Yes | The hyperparameters of the solvers are fixed for all experiments and their values are listed in Table 1. ... Table 1: Hyperparameters of DCOP solvers used during simulations. Algorithm Hyperparameters DPOP SD-Gibbs iterations = 20 DUCT ϵ = 0.6, δ = 0.1 AC-DPOP iterations = 100, δ = 0.001, α = 0.01 PFD particles = 2000, w = 0.9, c1 = 0.9, c2 = 0.1, maxfc = 5, maxsc = 15 D-Bay λ = Lf |