Mark R. Palmer

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.

Testing the robustness of management decisions to uncertainty: Everglades restoration scenarios.

To effectively manage large natural reserves, resource managers must prepare for future contingencies while balancing the often conflicting priorities of different stakeholders. To deal with these issues, managers routinely employ models to project the response of ecosystems to different scenarios that represent alternative management plans or environmental forecasts. Scenario analysis is often used to rank such alternatives to aid the decision making process. However, model projections are subject to uncertainty in assumptions about model structure, parameter values, environmental inputs, and subcomponent interactions. We introduce an approach for testing the robustness of model-based management decisions to the uncertainty inherent in complex ecological models and their inputs. We use relative assessment to quantify the relative impacts of uncertainty on scenario ranking. To illustrate our approach we consider uncertainty in parameter values and uncertainty in input data, with specific examples drawn from the Florida Everglades restoration project. Our examples focus on two alternative 30-year hydrologic management plans that were ranked according to their overall impacts on wildlife habitat potential. We tested the assumption that varying the parameter settings and inputs of habitat index models does not change the rank order of the hydrologic plans. We compared the average projected index of habitat potential for four endemic species and two wading-bird guilds to rank the plans, accounting for variations in parameter settings and water level inputs associated with hypothetical future climates. Indices of habitat potential were based on projections from spatially explicit models that are closely tied to hydrology. For the American alligator, the rank order of the hydrologic plans was unaffected by substantial variation in model parameters. By contrast, simulated major shifts in water levels led to reversals in the ranks of the hydrologic plans in 24.1-30.6% of the projections for the wading bird guilds and several individual species. By exposing the differential effects of uncertainty, relative assessment can help resource managers assess the robustness of scenario choice in model-based policy decisions.

Machine learning applications to job shop scheduling

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.

Credit assignment and discovery in classifier systems

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

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.

A grid service module for natural-resource managers

To facilitate transparent use of the high-performance Across Trophic-Level System Simulation (ATLSS) ecosystem-modeling package for natural-resource management, the authors developed a grid service module. The module exploits grid middleware functionality to process complex computation without requiring users to handle underlying issues. It represents the first application of grid computing to this discipline and provides a potential template for researchers in other disciplines to exploit scientific computation without extensive training in high-performance computing.

Discovering and Refining Algorithms Through Machine Learning

The development of solution methods for classes of operations research problems involves formulating the problems mathematically, developing algorithms to solve the abstracted formulations and evaluating the solutions. One possible contribution of artificial intelligence to this process is the application of machine learning to algorithm discovery and refinement. This paper presents several genetic algorithm and classifier system based experiments to discover and refine algorithms for simple scheduling problems. The discovered algorithms can be considered to be rule bases that are modified and adapted through training with examples. The quality of the resultant algorithm is investigated as a function of the training.