Machine learning methods for parameter tuning in heuristics

Parameter tuning is a crucial issue both in the scientific development and in the practical use of heuristics. In some cases the role of the user as an intelligent (learning) part makes the reproducibility of heuristic results difficult and the competitiveness of alternative techniques dependent in a crucial way on the user capabilities. We argue that some simple subsymbolic machine learning methods can be profitably used in order to automate the tuning process and make it an integral (and fully documented) part of the algorithm. If learning acts on-line, task-dependent local properties can be used by the algorithm to determine the appropriate balance between diversification and intensification. In this way a single algorithm maintains the flexibility to deal with related problems through an internal feedback loop that considers the previous history of the search. The practical success of the approach in some cases raises the need of a sounder theoretical foundation of non-Markovian search techniques.

[1]  Longxin Lin.  Self-Improving Reactive Agents Based on Reinforcement Learning, Planning and Teaching , 2004, Machine Learning.

[2]  C. D. Gelatt,et al.  Optimization by Simulated Annealing , 1983, Science.

[3]  Ching-Fang Liaw,et al.  A Bibliography of Heuristic Search Research Through 1992 , 1994, IEEE Trans. Syst. Man Cybern. Syst..

[4]  Christos H. Papadimitriou,et al.  Beyond competitive analysis [on-line algorithms] , 1994, Proceedings 35th Annual Symposium on Foundations of Computer Science.

[5]  Kenneth Steiglitz,et al.  Combinatorial Optimization: Algorithms and Complexity , 1981 .

[6]  Pierre Hansen,et al.  Algorithms for the maximum satisfiability problem , 1987, Computing.

[7]  Cecilia R. Aragon,et al.  Optimization by Simulated Annealing: An Experimental Evaluation; Part II, Graph Coloring and Number Partitioning , 1991, Oper. Res..

[8]  Umesh V. Vazirani,et al.  An Introduction to Computational Learning Theory , 1994 .

[9]  Fred W. Glover,et al.  Tabu Search - Part I , 1989, INFORMS J. Comput..

[10]  Roberto Battiti,et al.  The Reactive Tabu Search , 1994, INFORMS J. Comput..

[11]  Fred Glover,et al.  Tabu Search - Part II , 1989, INFORMS J. Comput..

[12]  T. Chiang,et al.  On the convergence rate of annealing processes , 1987 .

[13]  Jan Korst,et al.  Deterministic and randomized local search , 1993 .

[14]  Mauricio G. C. Resende,et al.  Designing and reporting on computational experiments with heuristic methods , 1995, J. Heuristics.

[15]  Leslie G. Valiant,et al.  A theory of the learnable , 1984, CACM.

[16]  AngluinDana.  Learning regular sets from queries and counterexamples , 1987 .

[17]  Afonso Ferreira,et al.  BOUNDING THE PROBABILITY OF SUCCESS OF STOCHASTIC METHODS FOR GLOBAL OPTIMIZATION , 1993 .

[18]  Philip N. Strenski,et al.  Analysis of finite length annealing schedules , 2005, Algorithmica.

[19]  Dana Angluin,et al.  Learning Regular Sets from Queries and Counterexamples , 1987, Inf. Comput..

[20]  Roberto Battiti,et al.  Reactive Local Search for the Maximum Clique Problem1 , 2001, Algorithmica.

[21]  Yaser S. Abu-Mostafa,et al.  The Vapnik-Chervonenkis Dimension: Information versus Complexity in Learning , 1989, Neural Computation.

[22]  David Haussler,et al.  What Size Net Gives Valid Generalization? , 1989, Neural Computation.