Weaving of Metaheuristics with Cooperative Parallelism

Abstract

The user has requested enhancement of the downloaded file.

A Permutational Boltzmann Machine with Parallel Tempering for Solving Combinatorial Optimization Problems

“…We compare the performance of our systems against eight state-of-the-art solvers, described in Table 1. Solvers [4,5,22] use a preset iteration/time limit as their termination criterion while [15,[18][19][20]24] terminate as soon as the BKS is reached. All metrics are taken directly from the respective papers.…”

“…Replicas are generally arranged in order of increasing T from T min to T max in a temperature ladder. A replica, R k , operating at temperature T k , can stochastically exchange temperature with the replica immediately above it on the ladder, R k+1 , with an Exchange Acceptance Probability (EAP ) calculated via (18). Figure 3 outlines the structure of an optimization engine using BM replicas with PT.…”

“…After all replicas have finished their Y trials, temperature exchanges are performed. Similar to QAP solvers such as [15,[18][19][20]24], this process is repeated until the state corresponding to the best-known-solution (BKS ) of a problem, E BKS , is reached by one of the replicas, terminating the loop. The system then returns, for each replica, the minimum energy found and the corresponding state.…”

Quadratic assignment problem variants: A survey and an effective parallel memetic iterated tabu search

“…Improvements were achived when RoTS, EO and a GA run in parallel using a framework designed for cooperative parallel local searches (CPLS-GA) in López et al [63]. While the CPLS framework is designed to handle single-solution metaheuristics, another framework was proposed to ease the implementation of hybrid metaheuristics using cooperative parallelism, called Parallel Hybridization of Simple Heuristics (PHYSH) in López et al [64]. Another metaheuristic hybridized with RoTS is the Artificial Bee Colony by Dokeroglu et al [28].…”

“…A version of PMITS without paralellism, or simply MITS, is compared to the sequential algorithms JRG-DivTS [54], ITS [77], TLBO-RTS (sequential) [26], and BMA [9] in Table 3. With parallelism on, PMITS is compared to CPTS [53], TLBO-RTS (parallel) [26], PHA [104], MSH-QAP [27], PHYSH-QAP [64], BLS-OpenMP [6], and PABC-QAP [28] in Table 4. Results are reported as the APD to the BKS, calculated as the average of AP D…”