Athanasios Kehagias

A Comparison of Word- and Sense-Based Text Categorization Using Several Classification Algorithms

Most of the text categorization algorithms in the literature represent documents as collections of words. An alternative which has not been sufficiently explored is the use of word meanings, also known as senses. In this paper, using several algorithms, we compare the categorization accuracy of classifiers based on words to that of classifiers based on senses. The document collection on which this comparison takes place is a subset of the annotated Brown Corpus semantic concordance. A series of experiments indicates that the use of senses does not result in any significant categorization improvement.

Efficient Multi-robot Search for a Moving Target

This paper examines the problem of locating a mobile, non-adversarial target in an indoor environment using multiple robotic searchers. One way to formulate this problem is to assume a known environment and choose searcher paths most likely to intersect with the path taken by the target. We refer to this as the multi-robot efficient search path planning (MESPP) problem. Such path planning problems are NP-hard, and optimal solutions typically scale exponentially in the number of searchers. We present an approximation algorithm that utilizes finite-horizon planning and implicit coordination to achieve linear scalability in the number of searchers. We prove that solving the MESPP problem requires maximizing a non-decreasing, submodular objective function, which leads to theoretical bounds on the performance of our approximation algorithm. We extend our analysis by considering the scenario where searchers are given noisy non-line-of-sight ranging measurements to the target. For this scenario, we derive and integrate online Bayesian measurement updating into our framework. We demonstrate the performance of our framework in two large-scale simulated environments, and we further validate our results using data from a novel ultra-wideband ranging sensor. Finally, we provide an analysis that demonstrates the relationship between MESPP and the intuitive average capture time metric. Results show that our proposed linearly scalable approximation algorithm generates searcher paths that are competitive with those generated by exponential algorithms.

Predictive modular neural networks for time series classification

A predictive modular neural network (PREMONN) architecture for time series classification is presented. The PREMONN has a hierarchical structure. The bottom level consists of a bank of linear or nonlinear predictor modules. The top level is a decision module which employs Bayesian or nonprobabilistic decision rules. For various choices of prediction and decision modules, convergence to correct classification is proven. Also it is shown that PREMONN is robust to noise and the speed/accuracy tradeoff is investigated. The analysis is mainly mathematical; however, we also present classification experiments to corroborate our conclusions. Copyright 1996 Elsevier Science Ltd.

A Dynamic Programming Algorithm for Linear Text Segmentation

In this paper we introduce a dynamic programming algorithm which performs linear text segmentation by global minimization of a segmentation cost function which incorporates two factors: (a) within-segment word similarity and (b) prior information about segment length. We evaluate segmentation accuracy of the algorithm by precision, recall and Beefermans segmentation metric. On a segmentation task which involves Chois text collection, the algorithm achieves the best segmentation accuracy so far reported in the literature. The algorithm also achieves high accuracy on a second task which involves previously unused texts.

A Bayesian Multiple Models Combination Method for Time Series Prediction

In this paper we present the Bayesian Combined Predictor (BCP), a probabilistically motivated predictor for time series prediction. BCP utilizes local predictors of several types (e.g., linear predictors, artificial neural network predictors, polynomial predictors etc.) and produces a final prediction which is a weighted combination of the local predictions; the weights can be interpreted as Bayesian posterior probabilities and are computed online. Two examples of the method are given, based on real world data: (a) short term load forecasting for the Greek Public Power Corporation dispatching center of the island of Crete, and (b) prediction of sugar beet yield based on data collected from the Greek Sugar Industry. In both cases, the BCP outperforms conventional predictors.

L -fuzzy valued inclusion measure, L -fuzzy similarity and L -fuzzy distance

The starting point of this paper is the introduction of a new measure of inclusion of fuzzy set A in fuzzy set B. Previously used inclusion measures take values in the interval [0, 1]; the inclusion measure proposed here takes values in a Boolean lattice. In other words, an inclusion is viewed as an L-fuzzy valued relation between fuzzy sets. This relation is reflexive, antisymmetric and transitive, i.e. it is a fuzzy order relation; in addition, it possesess a number of properties which various authors have postulated as axiomatically appropriate for an inclusion measure. We also define an L-fuzzy valued measure of similarity between fuzzy sets and an L-fuzzy valued distance function between fuzzy sets; these possess properties analogous to the ones of real-valued similarity and distance functions.

Time-series segmentation using predictive modular neural networks

A predictive modular neural network method is applied to the problem of unsupervised time-series segmentation. The method consists of the concurrent application of two algorithms: one for source identification, the other for time-series classification. The source identification algorithm discovers the sources generating the time series, assigns data to each source, and trains one predictor for each source. The classification algorithm recursively computes a credit function for each source, based on the competition of the respective predictors, according to their predictive accuracy; the credit function is used for classification of the time-series observation at each time step. The method is tested by numerical experiments.

Modular neural networks for MAP classification of time series and the partition algorithm

We apply the partition algorithm to the problem of time-series classification. We assume that the source that generates the time series belongs to a finite set of candidate sources. Classification is based on the computation of posterior probabilities. Prediction error is used to adaptively update the posterior probability of each source. The algorithm is implemented by a hierarchical, modular, recurrent network. The bottom (partition) level of the network consists of neural modules, each one trained to predict the output of one candidate source. The top (decision) level consists of a decision module, which computes posterior probabilities and classifies the time series to the source of maximum posterior probability. The classifier network is formed from the composition of the partition and decision levels. This method applies to deterministic as well as probabilistic time series. Source switching can also be accommodated. We give some examples of application to problems of signal detection, phoneme, and enzyme classification. In conclusion, the algorithm presented here gives a systematic method for the design of modular classification networks. The method can be extended by various choices of the partition and decision components.

Probabilistic Strategies for Pursuit in Cluttered Environments with Multiple Robots

In this paper, we describe a method for coordinating multiple robots in a pursuit-evasion domain. We examine the problem of multiple robotic pursuers attempting to locate a non-adversarial mobile evader in an indoor environment. Unlike many other approaches to this problem, our method seeks to minimize expected time of capture rather than guaranteeing capture. This allows us to examine the performance of our algorithm in complex and cluttered environments where guaranteed capture is difficult or impossible with limited pursuers. We present a probabilistic formulation of the problem, discretize the environment, and define cost heuristics for use in planning. We then propose a scalable algorithm using an entropy cost heuristic that searches possible movement paths to determine coordination strategies for the robotic pursuers. We present simulated results describing the performance of our algorithm against state of the art alternatives in a complex office environment. Our algorithm successfully reduces capture time with limited pursuers in an environment beyond the scope of many other approaches.

Novel Fuzzy Inference System (FIS) Analysis and Design Based on Lattice Theory

We introduce novel (set- and lattice-theoretic) perspectives and tools for the analysis and design of fuzzy inference systems (FISs). We present an FIS, including both fuzzification and defuzzification, as a device for implementing a function f: RNrarr RM. The family of FIS functions has cardinality aleph2=2aleph1, where aleph1 is the cardinality of the set R of real numbers. Hence the FIS family is much larger than polynomials, neural networks, etc.; furthermore a FIS has a capacity for local generalization. A formulation in the context of lattice theory allows us to define the set F* of fuzzy interval numbers (FINs), which includes both (fuzzy) numbers and intervals. We present a metric dK on F*, which can introduce tunable nonlinearities. FIS design based on dK has advantages such as: an alleviation of the curse of dimensionality problem and a potential for improved computer memory utilization. We present a new FIS classifier, namely granular self-organizing map (grSOM), which we apply to an industrial fertilizer modeling application