ABSTRACT
Stochastic local search (SLS) algorithms are among the most successful techniques for solving computationally hard problems from computing science, operations research and various application areas; examples range from propositional satisfiability, routing- and scheduling problems to genome sequence assembly, protein structure prediction and winner determination in combinatorial auctions. Among the components underlying any SLS algorithm, the neighbourhood relation and the step function are particularly important. Typically, neighbourhood relations have to be defined in a problem-specific way, and it is often difficult to predict which of the various choices that can be made in this context will result in the best performance. One of the most basic SLS methods is based on the idea of iteratively improving a candidate solution of the given problem with respect to an evaluation function. More precisely, the search is started from some initial position, and in each search step, the candidate solution s is replaced with a neighbouring candidate solution s' with better evaluation function value.
