Walter A. Kosters

Genetic Programming for data classification: partitioning the search space

When Genetic Programming is used to evolve decision trees for data classification, search spaces tend to become extremely large. We present several methods using techniques from the field of machine learning to refine and thereby reduce the search space sizes for decision tree evolvers. We will show that these refinement methods improve the classification performance of our algorithms.

Data Mining Approaches to Criminal Career Analysis

Narrative reports and criminal records are stored digitally across individual police departments, enabling the collection of this data to compile a nation-wide database of criminals and the crimes they committed. The compilation of this data through the last years presents new possibilities of analyzing criminal activity through time. Augmenting the traditional, more socially oriented, approach of behavioral study of these criminals and traditional statistics, data mining methods like clustering and prediction enable police forces to get a clearer picture of criminal careers. This allows officers to recognize crucial spots in changing criminal behaviour and deploy resources to prevent these careers from unfolding. Four important factors play a role in the analysis of criminal careers: crime nature, frequency, duration and severity. We describe a tool that extracts these from the database and creates digital profiles for all offenders. It compares all individuals on these profiles by a new distance measure and clusters them accordingly. This method yields a visual clustering of these criminal careers and enables the identification of classes of criminals. The proposed method allows for several user-defined parameters.

TETRIS IS HARD, EVEN TO APPROXIMATE

In the popular computer game of Tetris, the player is given a sequence of tetromino pieces and must pack them into a rectangular gameboard initially occupied by a given configuration of filled squares; any completely filled row of the gameboard is cleared and all filled squares above it drop by one row. We prove that in the offline version of Tetris, it is -complete to maximize the number of cleared rows, maximize the number of tetrises (quadruples of rows simultaneously filled and cleared), minimize the maximum height of an occupied square, or maximize the number of pieces placed before the game ends. We furthermore show the extreme inapproximability of the first and last of these objectives to within a factor of p1-e, when given a sequence of p pieces, and the inapproximability of the third objective to within a factor of 2-e, for any e>0. Our results hold under several variations on the rules of Tetris, including different models of rotation, limitations on player agility, and restricted piece sets.

Interesting Fuzzy Association Rules in Quantitative Databases

In this paper we examine association rules and their interestingness. Usually these rules are discussed in the world of basket analysis. Instead of customer data we now study the situation with data records of a more general but fixed nature, incorporating quantitative (nonboolean) data. We propose a method for finding interesting rules with the help of fuzzy techniques and taxonomies for the items/attributes. Experiments show that the use of the proposed interestingness measure substantially decreases the number of rules.

Mining Clusters with Association Rules

In this paper we propose a method for extracting clusters in a population of customers, where the only information available is the list of products bought by the individual clients. We use association rules having high confidence to construct a hierarchical sequence of clusters. A specific metric is introduced for measuring the quality of the resulting clusterings. Practical consequences are discussed in view of some experiments on real life datasets.

Formal models of appraisal: Theory, specification, and computational model

Cognitive appraisal theories (CATs) explain human emotions as a result of the subjective evaluation of events that occur in the environment. Recently, arguments have been put forward that discuss the need for formal descriptions in order to further advance the field of cognitive appraisal theory. Formal descriptions can provide detailed predictions and help to integrate different CATs by providing clear identification of the differences and similarities between theories. A computational model of emotion that is based on a CAT also needs formal descriptions specifying the theory on which it is based. In this paper we propose a formal notation for the declarative semantics of the structure of appraisal. We claim that this formalism facilitates both integration of appraisal theories as well as the design and evaluation of computational models of emotion based on an appraisal theory. To support these claims we show how our formalism can be used in both ways: first we integrate two appraisal theories; second, we use this formal integrated model as basis for a computational model after identifying what declarative information is missing in the formal model. Finally, we embed the computational model in an emotional agent, and show how the formal specification helps to evaluate the computational model.

Computing the Eccentricity Distribution of Large Graphs

The eccentricity of a node in a graph is defined as the length of a longest shortest path starting at that node. The eccentricity distribution over all nodes is a relevant descriptive property of the graph, and its extreme values allow the derivation of measures such as the radius, diameter, center and periphery of the graph. This paper describes two new methods for computing the eccentricity distribution of large graphs such as social networks, web graphs, biological networks and routing networks.We first propose an exact algorithm based on eccentricity lower and upper bounds, which achieves significant speedups compared to the straightforward algorithm when computing both the extreme values of the distribution as well as the eccentricity distribution as a whole. The second algorithm that we describe is a hybrid strategy that combines the exact approach with an efficient sampling technique in order to obtain an even larger speedup on the computation of the entire eccentricity distribution. We perform an extensive set of experiments on a number of large graphs in order to measure and compare the performance of our algorithms, and demonstrate how we can efficiently compute the eccentricity distribution of various large real-world graphs.

Complexity analysis of depth first and FP-growth implementations of APRIORI

We examine the complexity of Depth First and FP-growth implementations of Apriori, two of the fastest known data mining algorithms to find frequent itemsets in large databases. We describe the algorithms in a similar style, derive theoretical formulas, and provide experiments on both synthetic and real life data to illustrate the theory.

Determining the diameter of small world networks

In this paper we present a novel approach to determine the exact diameter (longest shortest path length) of large graphs, in particular of the nowadays frequently studied small world networks. Typical examples include social networks, gene networks, web graphs and internet topology networks. Due to complexity issues, the diameter is often calculated based on a sample of only a fraction of the nodes in the graph, or some approximation algorithm is applied. We instead propose an exact algorithm that uses various lower and upper bounds as well as effective node selection and pruning strategies in order to evaluate only the critical nodes which ultimately determine the diameter. We will show that our algorithm is able to quickly determine the exact diameter of various large datasets of small world networks with millions of nodes and hundreds of millions of links, whereas before only approximations could be given.

Affect, Anticipation, and Adaptation: Affect-Controlled Selection of Anticipatory Simulation in Artificial Adaptive Agents

Emotion plays an important role in thinking. In this article we study affective control of the amount of simulated anticipatory behavior in adaptive agents using a computational model. Our approach is based on model-based reinforcement learning (RL) and inspired by the simulation hypothesis (Cotterill, 2001; Hesslow, 2002). The simulation hypothesis states that thinking is internal simulation of behavior using the same sensory-motor systems as those used for overt behavior. Here, we study the adaptiveness of an artificial agent, when action-selection bias is induced by an affect-controlled amount of simulated anticipatory behavior . To this end, we introduce an affect-controlled simulation-selection mechanism that uses the predictions of the agents RL model to select anticipatory behaviors for simulation. Based on eXperiments with adaptive agents in two nondeterministic partially observable grid-worlds we conclude that (1) internal simulation has an adaptive benefit and (2) affective control can reduce the amount of simulation needed for this benefit. This is specifically the case if the following relation holds: positive affect decreases the amount of simulation towards simulating the best potential neXt action, while negative affect increases the amount of simulation towards simulating all potential neXt actions. In essence we use artificial affect to control mental eXploration versus eXploitations. Thus, agents “feeling positive” can think ahead in a narrow sense and free up working memory resources, while agents “feeling negative” must think ahead in a broad sense and maXimize usage of working memory. Our results are consistent with several psychological findings on the relation between affect and learning, and contribute to answering the question of when positive versus negative affect is useful during adaptation.