Gunar E. Liepins

Representational issues in genetic optimization

Abstract Functions are partially characterized as easy or hard for genetic algorithms to optimize. The failure modes of inappropriate embedding, crossover disruption, and deceptiveness are introduced, analyzed, and resolved in part. Virtually all optimizable (by any method) real valued functions defined on a finite domain are shown to be theoretically easy for genetic algorithms given appropriately chosen representations. Unfortunately, problems that are easy in theory can be difficult in practice because of sampling error. Also, the transformations required to induce favorable representations are generally arbitrary permutations, and the space of permutations is so large that search for good ones is intractable. The space of inversions is amenable to search, but inversions are insufficiently powerful to overcome deceptiveness. On the other hand, affine transformations (over the diadic group) are shown to be sufficiently powerful to transform at least selected deceptive problems into easy ones. These new ...

Genetic Algorithms Applications to Set Covering and Traveling Salesman Problems

For set covering problems, genetic algorithms with two types of crossover operators are investigated in conjunction with three penalty function and two multiobjective formulations. A Pareto multiobjective formulation and greedy crossover are suggested to work well. On the other hand, for traveling salesman problems, the results appear to be discouraging; genetic algorithm performance hardly exceeds that of a simple swapping rule. These results suggest that genetic algorithms have their place in optimization of constrained problems. However, lack of, or insufficient use of fundamental building blocks seems to keep the tested genetic algorithm variants from being competitive with specialized search algorithms on ordering problems.

Characterizing crossover in genetic algorithms

We characterize crossover and schemata; crossover is a binary operator that preserves schemata and commutes with addition and projection. Moreover, for any setS of chromosomes and familyF of crossover operators, we fully characterize the reachable chromosomes.

Data quality control theory and pragmatics

Data Quality Control: Theory and Pragmatics. Edited by G. E. Liepins and V.R.R. Uppuluri. ISBN 0 8247 8354 9. Dekker, New York, 1991. xii + 360 pp.

Optimal Imputation of Erroneous Data: Categorical Data, General Edits

107.50.

Machine learning applications to job shop scheduling

Responses to surveys often contain large amounts of incorrect information. One option for dealing with the problem is to revise those erroneous responses that can be detected. Fellegi and Holt developed a model in which a response is modified to pass a set of edits with as little change as possible. The model is called Minimum Weighted Fields to Impute MWFI and is NP-hard for categorical data and general edits. We develop two algorithms for MWFI, based on set covering, and present computational experience.

Credit assignment and discovery in classifier systems

Reactive scheduling is the determination of a satisfactory schedule for act iv i t ies whenever such decisions need to be made quickly and without the ab i l i ty to fu l ly simulate the events, usually in response to a malfunction or unexpected event. This paper advocates augmenting expertly known heuristics for react ive scheduling with heuristics discovered through machine learning. Machine learning techniques are applied to learn scheduling heuristics for simple job shop scheduling problems.

Diagnosis, parsimony, and genetic algorithms

Classifier systems are “discovery” production rule systems that utilize the genetic algorithm for discovery and allocate credit through the bucket brigade. For any given problem, the success of a classifier system depends on the choice of representation, the systems ability to attain reward or punishment states (evaluation states), accurate estimation of the relative merit of individual classifiers, and the genetic algorithms ability to use information about the current population of rules to generate better rules. This article addresses the adequacy of the bucket brigade and backward averaging for credit assignment and reviews a preliminary study of two variants in conjunction with rules that are fully enumerated as well as with discovery. Potential difficulties with each of these methods are highlighted in several theoretical examples, including one from the literature. Preliminary results and tentative similarities between these hybrids and Suttons Adaptive Heuristic Critic (AHC) are suggested.

Learning decision rules for scheduling problems: a classifier hybrid approach

The Communication Alarm Processor Expert System (CAP), developed at Oak Ridge National Laboratory for the Bonneville Power Administration, is a near real-time system that aids microwave communication system operators with interpreting the cause of large communication system problems [Purucker89]. Problems in the communications network are indicated by the real-time arrival of alarms at the central control site. CAP receives and processes these alarms, then presents the operator with a sorted list indicating the most probable cause (and location) generating the alarms. However, to achieve multiple problem diagnosis a diagnostic strategy is needed that: 1) satisfies the previously defined near real-time processing constraints, 2) “scales up” easily to handle large real-world applications (i.e., applications with more than 50 problems/components), and 3) gives the operator highly reliable information on the current status of the communications network. This paper describes recent successful results of our efforts to develop a general multiple problem (fault) diagnostic strategy that meets these requirements. The CAP system is currently being upgraded to incorporate this new strategy.

Global convergence of genetic algorithms

ABSTRACT A series of experiments to learn general rules for simple job shop scheduling tasks suggest that the classifier system may work best as a component of a larger system. Preliminary results demonstrate the systems ability to learn binary decision rules as a component of a sorting routine.