Comparison of Acceptance Criteria in Randomized Local Searches
One key component of stochastic local search algorithms is the acceptance criterion that determines whether a solution is accepted as the new current solution or it is discarded. One of the most studied local search algorithms is simulated annealing. It often uses the Metropolis condition as acceptance criterion, which always accepts equal or better quality solutions and worse ones with a probability that depends on the amount of worsening and a parameter called temperature. After the introduction of simulated annealing several other acceptance criteria have been introduced to replace the Metropolis condition, some being claimed to be simpler and better performing. In this article, we evaluate various such acceptance criteria from an experimental perspective. We first tune the numerical parameters of the algorithms using automatic algorithm configuration techniques for two test problems, the quadratic assignment problem and a permutation flowshop problem. Our experimental results show that, while results may differ depending on the specific problem, the Metropolis condition and the late acceptance hill climbing rule are among the choices that obtain the best results.
We acknowledge support from the COMEX project (P7/36) within the IAP Programme of the BelSPO. Thomas Stützle acknowledges support from the Belgian F.R.S.-FNRS, of which he is a senior research associate.
- 3.Burkard, R.E., Çela, E., Pardalos, P.M., Pitsoulis, L.S.: The quadratic assignment problem. In: Handbook of Combinatorial Optimization, vol. 2, pp. 241–338. Kluwer Academic Publishers (1998)Google Scholar
- 4.Burke, E.K., Bykov, Y.: The late acceptance hill-climbing heuristic. Technical report CSM-192, University of Stirling (2012)Google Scholar
- 13.Mascia, F., López-Ibáñez, M., Dubois-Lacoste, J., Stützle, T.: From grammars to parameters: automatic iterated greedy design for the permutation flow-shop problem with weighted tardiness. In: Nicosia, G., Pardalos, P. (eds.) LION 2013. LNCS, vol. 7997, pp. 321–334. Springer, Heidelberg (2013). https://doi.org/10.1007/978-3-642-44973-4_36 CrossRefGoogle Scholar