Analyzing the Performance of Local Search Algorithms Using Generalized Hill Climbing Algorithms

Abstract

Generalized hill climbing algorithms provide a well-defined framework to model how local search algorithms address intractable discrete optimization problems. Monte Carlo search, simulated annealing, threshold accepting, and tabu search, among others, can all be formulated as particular generalized hill climbing algorithms. Moreover, generalized hill climbing algorithms provide a structure for classifying and studying a large body of stochastic and deterministic local search algorithms. In particular, the generalized hill climbing algorithm framework can be used to develop a general Markov chain model for the application of local search algorithms to intractable discrete optimization problems. Moreover, the generalized hill climbing algorithm framework has been used to evaluate the finite-time performance of local search algorithms. This analysis is of particular interest to practitioners for whom traditional convergence results are often of limited interest and value. This paper presents a survey of recent results on how the generalized hill climbing algorithm framework has been used to model and gain insight into the performance of local search algorithms. Many of these results provide tools for comparing and evaluating different types of local search algorithms using common performance measures.KeywordsLocal SearchSimulated AnnealingTabu SearchSimulated Annealing AlgorithmLocal Search AlgorithmThese keywords were added by machine and not by the authors. This process is experimental and the keywords may be updated as the learning algorithm improves.