State-space search redundancy, that is, multiple explorations of the same state, is an inherent problem in many heuristic search algorithms. It is prevalent in constructive multi-start algorithms. Record-keeping mechanisms, however, can minimize redundancy and enable exploiting time/space tradeoffs. This paper investigates the utility of record-keeping procedures in the context of Iterative Hill Climbing applied to the Traveling Salesperson Problem using several restart mechanisms including Greedy Randomized Adaptive Search, and Greedy Enumeration. Record-keeping methods such as unbounded memory, dedicated memory, and cache memory, as well as a novel "book-keeping" method utilizing a Bloom filter are investigated. Experiments performed using TSPLIB benchmarks and random TSP instances with 100 cities show that under the above mentioned restart and record-keeping mechanisms the IHC produces competitive results. In addition, the research shows that record-keeping, in specific Bloom filters, can considerably improve both the time performance of IHC and the quality of solutions produced. © 2012 Springer-Verlag.
CITATION STYLE
Tamir, D. E., King, C. R., & McKenney, M. (2012). Improving the performance of constructive multi-start search using record-keeping. In Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) (Vol. 7345 LNAI, pp. 176–186). https://doi.org/10.1007/978-3-642-31087-4_19
Mendeley helps you to discover research relevant for your work.