This paper describes algorithms that learn to improve search performance on largescale optimization tasks. The main algorithm, Stage, works by learning an evaluation function that predicts the outcome of a local search algorithm, such as hillclimbing or Walksat, from features of states visited during search. The learned evaluation function is then used to bias future search trajectories toward better optima on the same problem. Another algorithm, X-Stage, transfers previously learned evaluation functions to new, similar optimization problems. Empirical results are provided on seven large-scale optimization domains: bin-packing, channel routing, Bayesian network structure-finding, radiotherapy treatment planning, cartogram design, Boolean satisfiability, and Boggle board setup.

This paper describes Stage, a learning algorithm that automatically improves search performance on large-scale optimization problems. Stage learns an evaluation function that predicts the outcome of a local search algorithm, such as hillclimbing or Walksat, from features of states visited during search. The learned evaluation function is used to bias future search trajectories toward better optima on the same problem. This paper presents the Stage algorithm; an extension, X-Stage, that transfers learned evaluation functions to new, similar optimization problems; and empirical results on seven large-scale optimization domains: bin-packing, channel routing, Bayes network structure-finding, radiotherapy treatment planning, cartogram design, Boolean satisfiability, and Boggle board setup.

There are many people I would like to thank for their support. Principally I thank my family, Sue, Nick, Katie and William, and my Supervisor Larry Bull. All have shown tolerance, forbearance, support and inspiration well beyond the call of duty. In addition Dave Wyatt and Praminda Caleb-Solly have been very knowledgeable, approachable and helpful, I owe them a lot. The UWE LCSG has been a key source of ideas, knowledge and inspiration. Books and papers cannot provide such a level of understanding and debate about fundamental issues. Amongst the key players Larry Bull, Alwyn Barry, Dave Wyatt, Matt Studley, Chris Stone, Tony Pipe, Brian Carse and Rob Smith have always provided difficult and thought provoking questions, and sometimes even answers. Lastly thanks to Paul Lewis, Joe Mackenzie both of BT, Terry Fogarty, and Roger Miles who in there own ways facilitated the move from being a competent project manager to entering the (from Huxley) “brave new world that hath such people in’t ” of evolutionary and neural computing. This thesis investigates a hybrid of evolutionary computing and neural computing which long has been a goal of machine learning. X-NCS is a neural and hence a more complex version of XCS (Wilson 1995), the pre-eminent accuracy based Learning Classifier System (LCS) (Holland, 1986). XCS differs from other