Queued Pareto Local Search for Multi-Objective Optimization

Abstract

General rightsIt is not permitted to download or to forward/distribute the text or part of it without the consent of the author(s) and/or copyright holder(s), other than for strictly personal, individual use, unless the work is under an open content license (like Creative Commons). Disclaimer/Complaints regulationsIf you believe that digital publication of certain material infringes any of your rights or (privacy) interests, please let the Library know, stating your reasons. In case of a legitimate complaint, the Library will make the material inaccessible and/or remove it from the website. Please Ask the Library: http://uba.uva.nl/en/contact, or a letter to: Library of the University of Amsterdam, Secretariat, Singel 425, 1012 WP Amsterdam, The Netherlands. You will be contacted as soon as possible. Abstract. Many real-world optimization problems involve balancing multiple objectives. When there is no solution that is best with respect to all objectives, it is often desirable to compute the Pareto front. This paper proposes queued Pareto local search (QPLS), which improves on existing Pareto local search (PLS) methods by maintaining a queue of improvements preventing premature exclusion of dominated solutions. We prove that QPLS terminates and show that it can be embedded in a genetic search scheme that improves the approximate Pareto front with every iteration. We also show that QPLS produces good approximations faster, and leads to better approximations than popular alternative MOEAs.

Distributed Pareto Local Search for Multi-Objective DCOPs

“…Similarly, exploration and restart strategies should be implemented in distributed settings as they proved to be quite successful for the centralized PLS [24]. Another method that should be studied for DPLS is the Queued PLS [25] that delays the removal of dominated solution from the archive as it can improve the approximation found by the algorithm. Furthermore, we intend to apply DPLS to challenging real world problems, e.g., sensor network and scheduling problems.…”

Multi-objective multi-agent decision making: a utility-based analysis and survey

“…Examples of such methods are multi-objective bucket elimination (MOBE) [94,93], also known as multi-objective variable elimination (MOVE, which is the more common in the planning and reinforcement learning communities), which solves a series of local sub-problems to eliminate all agents from a MOCoG in sequence, by finding local coverage sets as best responses to neighbouring agents. Other such methods include multi-objective Russian doll search [95], multi-objective (AND/OR) branch-andbound tree search [65,66,96] using mini-bucket heurtistics [94,65], Pareto local search [44], and multi-objective max-sum [22]. Many of these papers note that PCSs can grow very large very quickly, making finding exact PCSs infeasible.…”

PPLS/D: Parallel Pareto Local Search Based on Decomposition

“…Recently, several sequential PLS variants have been proposed to speed up the basic PLS. Inja et al [3] proposed the Queued PLS (QPLS). In QPLS a queue of high-quality unsuccessful candidate solutions are Algorithm 1 Pareto Local Search 1: input: An initial set of non-dominated solutions A 0 2: ∀x ∈ A 0 , set Explored(x) ← FALSE 3: A ← A 0 4: while A 0 = ∅ do 5:…”

“…Recently, several sequential PLS variants have been proposed to speed up the basic PLS. Inja et al [3] proposed the Queued PLS (QPLS). In QPLS a queue of high-quality unsuccessful candidate solutions are…”