Hybrid DCOP Solvers: Boosting Performance of Local Search Algorithms
This addresses efficiency issues for researchers and practitioners using DCOP algorithms, but it is incremental as it builds on existing solvers with a new initialization method.
The paper tackles the problem of slow convergence and high communication overhead in Distributed Constraint Optimization Problem (DCOP) solvers by initializing them with greedy fast non-iterative solvers instead of random assignments, resulting in up to 50% reduction in convergence time and improved solution quality.
We propose a novel method for expediting both symmetric and asymmetric Distributed Constraint Optimization Problem (DCOP) solvers. The core idea is based on initializing DCOP solvers with greedy fast non-iterative DCOP solvers. This is contrary to existing methods where initialization is always achieved using a random value assignment. We empirically show that changing the starting conditions of existing DCOP solvers not only reduces the algorithm convergence time by up to 50\%, but also reduces the communication overhead and leads to a better solution quality. We show that this effect is due to structural improvements in the variable assignment, which is caused by the spreading pattern of DCOP algorithm activation.) /Subject (Hybrid DCOPs)