In the feature subset selection problem, a learning algorithm is faced with the problem of selecting a relevant subset of features upon which to focus its attention, while ignoring the rest. To achieve the best possible performance with a particular learning algorithm on a particular training set, a feature subset selection method should consider how the algorithm and the training set interact. We explore the relation between optimal feature subset selection and relevance. Our wrapper method searches for an optimal feature subset tailored to a particular algorithm and a domain. We study the strengths and weaknesses of the wrapper approach andshow a series of improved designs. We compare the wrapper approach to induction without feature subset selection and to Relief, a filter approach to feature subset selection. Significant improvement in accuracy is achieved for some datasets for the two families of induction algorithms used: decision trees and Naive-Bayes.

. On-line learning in domains where the target concept depends on some hidden context poses serious problems. A changing context can induce changes in the target concepts, producing what is known as concept drift. We describe a family of learning algorithms that flexibly react to concept drift and can take advantage of situations where contexts reappear. The general approach underlying all these algorithms consists of (1) keeping only a window of currently trusted examples and hypotheses; (2) storing concept descriptions and re-using them when a previous context reappears; and (3) controlling both of these functions by a heuristic that constantly monitors the system's behavior. The paper reports on experiments that test the systems' performance under various conditions such as different levels of noise and different extent and rate of concept drift. Keywords: Incremental concept learning, on-line learning, context dependence, concept drift, forgetting 1. Introduction The work presen...

. We evaluate the power of decision tables as a hypothesis space for supervised learning algorithms. Decision tables are one of the simplest hypothesis spaces possible, and usually they are easy to understand. Experimental results show that on artificial and real-world domains containing only discrete features, IDTM, an algorithm inducing decision tables, can sometimes outperform state-of-the-art algorithms such as C4.5. Surprisingly, performance is quite good on some datasets with continuous features, indicating that many datasets used in machine learning either do not require these features, or that these features have few values. We also describe an incremental method for performing crossvalidation that is applicable to incremental learning algorithms including IDTM. Using incremental cross-validation, it is possible to cross-validate a given dataset and IDTM in time that is linear in the number of instances, the number of features, and the number of label values. The time for incre...

With the development of the semantic web activity, ontologies become widely used to represent the conceptualization of a domain. However, none of the existing ontology languages provides a means to capture uncertainty about the concepts, properties and instances in a domain. Probability theory is a natural choice for dealing with uncertainty. Incorporating probability theory into existing ontology languages will provide these languages additional expressive power to quantify the degree of the overlap or inclusion between two concepts, support probabilistic queries such as finding the most probable concept that a given description belongs to, and make more accurate semantic integration possible. One approach to provide such a probabilistic extension to ontology languages is to use Bayesian networks, a widely used graphic model for knowledge representation under uncertainty. In this paper, we present our on-going research on extending OWL, an ontology language recently proposed by W3C’s Semantic Web Activity. First, the language is augmented to allow additional probabilistic markups, so probabilities can be attached with individual concepts and properties in an OWL ontology. Secondly, a set of translation rules is defined to convert this probabilistically annotated OWL ontology into a Bayesian network. Our probabilistic extension to OWL has clear semantics: the Bayesian network obtained will be associated with a joint probability distribution over the application domain. General Bayesian network inference procedures (e.g., belief propagation or junction tree) can be used to compute P(C | e): the degree of the overlap or inclusion between a concept C and a concept represented by a description e. We also provide a similarity measure that can be used to find the most probable concept that a given description belongs to. 1.

Abstract not found

Abstract In this article, we present some extensions of the rough set approach and we outline a challenge for the rough set based research.

Abstract: This paper has two purposes. One is to present a critical examination of the rise of granular computing and the other is to suggest a triarchic theory of granular computing. By examining the reasons, justifications, and motivations for the rise of granular computing, we may be able to fully appreciate its scope, goal and potential values. The results enable us to formulate a triarchic theory in the light of research results from many disciplines. The three components of the theory are labeled as the philosophy, the methodology, and the computation. The integration of the three offers a unified view of granular computing as a way of structured thinking, a method of structured problem solving, and a paradigm of structured information processing, focusing on hierarchical granular structures. The triarchic theory is an important effort in synthesizing the various theories and models of granular computing. Key words: Triarchic theory of granular computing; systems theory; structured thinking, problem solving and information processing. CLC number: Document code: A Introduction Although granular computing, as a separate field of study, started a decade ago [1], its basic philosophy, ideas, principles, methodologies, theories and tools has, in fact, long been used either explicitly or implicitly across many branches of natural and social sciences The answers, at least partial answers, to these questions may be obtained by drawing and synthesizing results from well-established disciplines, including philosophy, psychology, neuroscience, cognitive science, education, artificial intelligence, computer programming, and many more. Previously, I argued that granular computing represents an idea converged from many branches of natural and social sciences Human-Inspired Computing Research on understanding the human brain and natural intelligence is closely related to the field of artificial intelligence (AI) and information technology (IT). The results have led to a computational view for explaining how the mind works

Abstract not found

Abstract. Real-life data usually are presented in databases by real numbers. On the other hand, most inductive learning methods require small number of attribute values. Thus it is necessary to convert input data sets with continuous attributes into input data sets with discrete attributes. Methods of discretization restricted to single continuous attributes will be called local, while methods that simultaneously convert all continuous attributes will be called global. In this paper, a method of transforming any local discretization method into a global one is presented. A global discretization method, based on cluster analysis, is presented and compared experimentally with three known local methods, transformed into global. Experiments include ten-fold cross validation and leaving-one-out methods for ten real-life data sets.

Information granulation and concept approximation are some of the fundamental issues of granular computing. Granulation of a universe involves grouping of similar elements into granules to form coarse-grained views of the universe. Approximation of concepts, represented by subsets of the universe, deals with the descriptions of concepts using granules. In the context of rough set theory, this paper examines the two related issues. The granulation structures used by standard rough set theory and the corresponding approximation structures are reviewed. Hierarchical granulation and approximation structures are studied, which results in stratified rough set approximations. A nested sequence of granulations induced by a set of nested equivalence relations leads to a nested sequence of rough set approximations. A multi-level granulation, characterized by a special class of equivalence relations, leads to a more general approximation structure. The notion of neighborhood systems is also explored. 1