Jacek Jelonek

ROUGH SET REDUCTION OF ATTRIBUTES AND THEIR DOMAINS FOR NEURAL NETWORKS

This paper presents an empirical study of the use of the rough set approach to reduction of data for a neural network classifying objects described by quantitative and qualitative attributes. Two kinds of reduction are considered: reduction of the set of attributes and reduction of the domains of attributes. Computational tests were performed with five data sets having different character, for original and two reduced representations of data. The learning time acceleration due to data reduction is up to 4.72 times. The resulting increase of misclassification error does not exceed 11.06%. These promising results let us claim that the rough set approach is a useful tool for preprocessing of data for neural networks.

Feature subset selection for classification of histological images

Classification of histological images is considered in this paper. The task is to distinguish different classes of tumours of the central nervous system on the basis of features extracted from microscopic slides. The number of extracted features is relatively high and some of them seem to be irrelevant for classification of the images. Thus, the main objective of this study is to select such a feature subset that improves the predictive accuracy of the classifier. The wrapper approach is chosen to obtain this aim, where a search for the good subset of features is made using a non-parametric case-base classifier. To guide a search process, a forward beam selection algorithm is introduced. It sequentially adds relevant features in a parallel way for the most promising subsets. It is shown that the proposed approach gives good predictive accuracy for the considered histopathological problem.

Experiments on Solving Multiclass Learning Problems by n2-classifier

The paper presents an experimental study of solving multiclass learning problems by a method called n2-classifier. This approach is based on training (n2 - n)/2 binary classifiers - one for each pair of classes. Final decision is obtained by a weighted majority voting rule. The aim of the computational experiment is to examine the influence of the choice of a learning algorithm on a classification performance of the n2-classifier. Three different algorithms are n2-classifier. decision trees, neural networks and instance based learning algorithm.

Neural Networks and Rough Sets - Comparison and Combination for Classification of Histological Pictures

Learning neural networks using large sets of data often appears to be very complex task. Well-known backpropagation algorithm for the learning of non-linear layered networks requires a lot of floating-point computation. The learning time is proportional to (among others) the size of the input vectors, so any reasonable reduction of the redundant information on the input is welcome. Rough set approach to reduction of the input data to a neural network is presented in this paper. Over two hundred records describing microscopic slides of 7 classes of brain tumours constitute considered data set.

Biologically-inspired visual-motor coordination model in a navigation problem

This work presents a biologically-inspired coordination model which associates motor actions with visual stimuli. The model is introduced and explained, and navigation experiments are reported that verify the implemented visual-motor system. Experiments demonstrate that the system can be trained to solve navigation problems consisting in moving around a 3D object to reach a specific location based on the visual information only. The model is flexible, as it is composed of an adjustable number of modules. It is also interpretable, i.e. it is possible to estimate the influence of visual features on the motor action.

Psychomotor and executive functioning after unilateral posteroventral pallidotomy in patients with Parkinson's disease

The goal of the study was to explore the immediate effects of unilateral posteroventral stereotactic pallidotomy (PVP) on psychomotor and executive functioning in patients with Parkinsons disease (PD). The original drawing task, conducted on a digitizing tablet, and neuropsychological tests were administered to 25 patients with PD, 2 or 3 days before and after the surgery. To assess executive functions, the following tests were used: Trail Making Test (TMT), Stroop Colour Interference Test and Wisconsin Card Sorting Test (WCST). To evaluate global mental functioning, Mini Mental State Examination (MMSE) was applied. Benton Visual Retention Test (BVRT) was introduced as a control non-executive task. The patients undergoing a surgery were compared with age and education matched healthy and PD controls. PVP resulted in an increased movability of the upper contralateral limbs reflected in larger average pressure put during the drawing task after the surgery. Assessment of the emotional state showed a significant postoperative improvement. An isolated significant decline of WCST performance, not related to the side of the lesion, was observed immediately after the surgery. The performance of the other executive and non-executive tasks remained unchanged. The results showed that unilateral PVP may lead to immediate selective executive impairment and is needed to be explored in further studies.

Inducing Jury’s Preferences in Terms of Acoustic Features of Violin Sounds

A set of violins submitted to a competition has been evaluated by the jury from the viewpoint of several criteria and then ranked from the best to the worst. The sound of the instruments played by violinists during the competition has been recorded digitally and then processed to obtain sound attributes. Given the jury’s ranking of violins according to sound quality criteria, we are inferring from the sound characteristics a preference model of the jury in the form of “if..., then...” decision rules. This preference model explains the given ranking and permits to build a ranking of a new set of violins according to this policy. The inference follows the scheme of an inductive supervised learning. For this, we are applying a special computational tool called Dominance-based Rough Set Approach (DRSA). The new set of attributes derived from the energy of consecutive halftones of the chromatic scales played on four strings has proved a good accuracy of the approximation.

Genetic Algorithms in Constructive Induction

In this paper, genetic algorithms are used in machine learning classification task. They act as a constructive induction engine, which selects features and adjusts weights of attributes, in order to obtain the highest classification accuracy. We compare two classification approaches: a single 1-NN and a n 2 meta-classifier. For the n 2-classifier, the idea of an improvement of classification accuracy is based on independent modification of descriptions of examples for each pair of n classes. Finally, it gives (n 2−n)/2 spaces of attributes dedicated for discrimination of pairs of classes. The computational experiment was performed on a medical data set. Its results reveal the utility of using genetic algorithms for features selection and weight adjusting, and point out the advantage of using a multi-classification model (n 2-classifier) with constructive induction in relation to the analogous single-classifier approach.