We present a novel framework for exploring very large state spaces of concurrent reactive systems. Our framework exploits application-independent heuristics using genetic algorithms to guide a state-space search towards error states. We have implemented this framework in conjunction with VeriSoft, a tool for exploring the state spaces of software applications composed of several concurrent processes executing arbitrary code. We present experimental results obtained with several examples of programs, including a C implementation of a public-key authentication protocol. We discuss heuristics and properties of state spaces that help a genetic search detect deadlocks and assertion violations. For nding errors in very large state spaces, our experiments show that a genetic search using simple heuristics can significantly outperform random and systematic searches.

In this paper we report the results of a comparative study on different variations of genetic programming applied on binary data classification problems. The first genetic programming variant is weighting data records for calculating the classification error and modifying the weights during the run. Hereby the algorithm is defining its own fitness function in an on-line fashion giving higher weights to `hard' records. Another novel feature we study is the atomic representation, where `Booleanization' of data is not performed at the root, but at the leafs of the trees and only Boolean functions are used in the trees' body. As a third aspect we look at generational and steady-state models in combination of both features.

Evolutionary algorithms (EAs) for solving constraint satisfaction problems (CSPs) can be roughly divided into two classes: EAs using adaptive fitness functions and EAs using heuristics. In [5] the most effective EAs of the first class have been compared experimentally using a large set of benchmark instances consisting of randomly generated binary CSPs. In this paper we complete this comparison by studying the most effective EAs that use heuristics.

In this chapter we describe a problem independent method for treating constraints in an evolutionary algorithm. Technically, this method amounts to changing the definition of the fitness function during a run of an EA, based on feedback from the search process. Obviously, redefining the fitness function means rede ning the problem to be solved. On the short term this deceives the algorithm making the fitness values deteriorate, but as experiments clearly indicate, on the long run it is beneficial. We illustrate the power of the method on different constraint satisfaction problems and point out other application areas of this technique.

We use evolutionary computation (EC) to automatically find problems which demonstrate the strength and weaknesses of modern search heuristics. In particular, we analyze particle swarm optimization (PSO), differential evolution (DE), and covariance matrix adaptation-evolution strategy (CMA-ES). Each evolutionary algorithm is contrasted with the others and with a robust nonstochastic gradient follower (i.e., a hill climber) based on Newton–Raphson. The evolved benchmark problems yield insights into the operation of PSOs, illustrate benefits and drawbacks of different population sizes, velocity limits, and constriction (friction) coefficients. The fitness landscapes made by genetic programming reveal new swarm phenomena, such as deception, thereby explaining how they work and allowing us to devise better extended particle swarm systems. The method could be applied to any type of optimizer.

In this tutorial we consider the issue of constraint handling by evolutionary algorithms (EA). We start this study with a categorization of constrained problems and observe that constraint handling is not straightforward in an EA. Namely, the search operators mutation and recombination are `blind' to constraints. Hence, there is no guarantee that if the parents satisfy some constraints the offspring will satisfy them as well. This suggests that the presence of constraints in a problem makes EAs intrinsically unsuited to solve this problem. This should especially hold if there are no objectives only constraints in the original problem specification -- the category of constraint satisfaction problems. A survey of related literature, however, discloses that there are quite a few successful attempts to evolutionary constraint satisfaction. Based on this survey we identify a number of common features in these approaches and arrive to the conclusion that the presence of constraints is not harmful, but rather helpful in that it provides extra information that EAs can utilize. The tutorial is concluded by considering a number of key questions on research methodology and some promising future research directions.

This paper introduces a genetic local search algorithm for binary constraint satisfaction problems. The core of the algorithm consists of an ad-hoc optimization procedure followed by the application of blind genetic operators. A standard set of benchmark instances is used in order to assess the performance of the algorithm. The results indicate that this apparently naive hybridation of a genetic algorithm with local search yields a rather powerful heuristic algorithm for random binary constraint satisfaction problems. Categories and Subject Descriptors G.1.6 [Mathematics of Computing]: Optimization---Global Optimization; I.2.8 [Artificial Intelligence]: Problem Solving, Control Methods, and Search---Heuristic methods General Terms Algorithms, Experimentation 1. INTRODUCTION In the binary constraint satisfaction problem (BCSP) we are given a set of variables, where each variable has a domain of values, and a set of constraints acting between pairs of variables. The problem consi...

Abstract. Constraint Satisfaction Problems form a class of problems that are generally computationally difficult and have been addressed with many complete and heuristic algorithms. We present two complete algorithms, as well as two evolutionary algorithms, and compare them on randomly generated instances of binary constraint satisfaction problems. We find that the evolutionary algorithms are less effective than the classical techniques. 1

Abstract. In this paper we describe how the Stepwise Adaptation of Weights (saw) technique can be applied in genetic programming. The saw-ing mechanism has been originally developed for and successfully used in eas for constraint satisfaction problems. Here we identify the very basic underlying ideas behind saw-ing and point out how it can be used for different types of problems. In particular, saw-ing is wellsuited for data mining tasks where the fitness of a candidate solution is composed by ‘local scores ’ on data records. We evaluate the power of the saw-ing mechanism on a number of benchmark classification data sets. The results indicate that extending the gp with the saw-ing feature increases its performance when different types of misclassifications are not weighted differently, but leads to worse results when they are. 1

This report is based on the work I have done for my Master Thesis project. The project as a whole consists of research done in the field of evolutionary computation, and it is split into two distinct parts. The main theme is adaptive evolutionary algorithms. The first part covers the research done on solving binary constraint satisfaction problems using adaptive evolutionary algorithms. This involves a comparative study on three algorithms, each of which incorporates a different adaptive fitness measure to guide its search to a solution for an instance of a binary constraint satisfaction problem. The second part mainly consists of the development of a library. Its use is aimed at evolutionary algorithms in general. Furthermore, a genetic programming algorithm is contructed, that incorporates an adaptive fitness measure. This construction served as a test of the usability of the library. The genetic programming algorithm has been used for experiments on different data sets from the data...