Stochastic local search (SLS) algorithms, like many other heuristic methods for solving hard
computational problems, are often built in an incremental way, guided by the experience and intuition
of the algorithm designer. This is particularly the case of relatively complex SLS methods, such 
as evolutionary algorithms or hybrid approaches that combine various simpler search strategies.
As a result, effective SLS strategies for many academic and real-world problems often consist
of many interactingcomponents, whose behaviour and interaction is controlled by numerous parameters.
In this talk, I will discuss strategies for automatically configuring these components and for
tuning their parameters in order to better realise the overall performance potential of the complex
algorithm. I will describe ParamILS, a recent, SLS-based algorithm configuration procedure developed 
in my group, that we have applied very successfully to a number of high-performing heuristic 
search methods, ranging from SLS algorithms to SAT to the widely used industrial 
mixed-integer-programming solver CPLEX. I will also review a formal model of hybrid SLS algorithms, 
Generalised Local Search Machines, and explain how, in conjunction with automated configuration 
methods, this model can provide the basis for the principled and automated synthesis of effective 
hybrid SLS algorithms.