Wyniki wyszukiwania - BazTech

1

Rule extraction from active contour classifiers

Szczepaniak P. S., Pierścieniewski Ł.

Journal of Applied Computer Science

|

2018

|

Vol. 26, nr 1

73--86

EN

In this paper, the idea of rule extraciton from active contour classifiers is presented. The concepts are new in relation to active contour approach. The problem is illustrated by examples having roots in technical diagnosis and in analysis of content of images.

2

Characterization of symbolic rules embedded in deep DIMLP networks : a challenge to transparency of deep learning

Bologna G., Hayashi Y.

Journal of Artificial Intelligence and Soft Computing Research

|

2017

|

Vol. 7, No. 4

265--286

EN

Rule extraction from neural networks is a fervent research topic. In the last 20 years many authors presented a number of techniques showing how to extract symbolic rules from Multi Layer Perceptrons (MLPs). Nevertheless, very few were related to ensembles of neural networks and even less for networks trained by deep learning. On several datasets we performed rule extraction from ensembles of Discretized Interpretable Multi Layer Perceptrons (DIMLP), and DIMLPs trained by deep learning. The results obtained on the Thyroid dataset and the Wisconsin Breast Cancer dataset show that the predictive accuracy of the extracted rules compare very favorably with respect to state of the art results. Finally, in the last classification problem on digit recognition, generated rules from the MNIST dataset can be viewed as discriminatory features in particular digit areas. Qualitatively, with respect to rule complexity in terms of number of generated rules and number of antecedents per rule, deep DIMLPs and DIMLPs trained by arcing give similar results on a binary classification problem involving digits 5 and 8. On the whole MNIST problem we showed that it is possible to determine the feature detectors created by neural networks and also that the complexity of the extracted rulesets can be well balanced between accuracy and interpretability.

3

Ant-based extraction of rules in simple decision systems over ontological graphs

Pancerz K., Lewicki A., Tadeusiewicz R.

International Journal of Applied Mathematics and Computer Science

|

2015

|

Vol. 25, no. 2

377--387

EN

In the paper, the problem of extraction of complex decision rules in simple decision systems over ontological graphs is considered. The extracted rules are consistent with the dominance principle similar to that applied in the dominance-based rough set approach (DRSA). In our study, we propose to use a heuristic algorithm, utilizing the ant-based clustering approach, searching the semantic spaces of concepts presented by means of ontological graphs. Concepts included in the semantic spaces are values of attributes describing objects in simple decision systems.

4

Discovering knowledge with the rough set approach

Mazurek J.

Polish Journal of Management Studies

|

2013

|

Vol. 7

245--254

EN

The rough set theory, which originated in the early 1980s, provides an alternative approach to the fuzzy set theory, when dealing with uncertainty, vagueness or inconsistence often encountered in real-world situations. The fundamental premise of the rough set theory is that every object of the universe is associated with some information, which is frequently imprecise and insufficient to distinguish among objects. In the rough set theory, this information about objects is represented by an information system (decision table). From an information system many useful facts and decision rules can be extracted, which is referred as knowledge discovery, and it is successfully applied in many fields including data mining, artificial intelligence learning or financial investment. The aim of the article is to show how hidden knowledge in the real-world data can be discovered within the rough set theory framework. After a brief preview of the rough set theory’s basic concepts, knowledge discovery is demonstrated on an example of baby car seats evaluation. For a decision rule extraction, the procedure of Ziarko and Shan is used.

PL

Teoria zbiorów przybliżonych, która powstała w roku 1980, oferuje alternatywne podejście do teorii zbiorów rozmytych, gdy ma się do czynienia ze zjawiskiem niepewności, niejasności i niekonsekwencji, często spotykanym w rzeczywistych sytuacjach. Podstawowym założeniem teorii zbiorów przybliżonych jest to, że każdy obiekt wszechświata jest związany z pewnymi informacjami, które są często nieprecyzyjne i niewystarczające do rozróżnienia między obiektami. W teorii zbiorów przybliżonych, informacje o obiektach są reprezentowane przez system informacyjny (tabela decyzyjna). System informacyjny dostarcza wiele przydatnych faktów i reguł, które są określane jako odkrywanie wiedzy, która z powodzeniem jest stosowana w wielu dziedzinach, w tym w ekstrakcji danych, sztucznej inteligencji czy przy inwestycjach finansowych. Cele artykułu jest pokazanie, w jaki sposób wiedza ukryta w rzeczywistych danych, mogą zostać odkryte w trudnych ramach teorii mnogości. Po krótkim przedstawieniu podstawowych pojęć teorii zbiorów przybliżonych, na przykładzie ocen fotelików samochodowych, przedstawiono zjawisko odkrywania wiedzy. W celu wydobycia reguły decyzyjnej zastosowano procedurę Ziarko i Shan.

5

Machine translation using scarce bilingual corpora

Jassem K., Kowalski T.

TASK Quarterly : scientific bulletin of Academic Computer Centre in Gdansk

|

2007

|

Vol. 11, No 1-2

21-33

EN

We propose a method for automatic extraction of translation rules suitable for a rule-based machine translation system by using a target language syntactic parser and scarce bilingual resources as linguistic knowledge sources. We propose an algorithm that assembles translation rules in order to translate an input sentence.

6

Rule extraction from approximating neural network - a real challenge for evolutionary method

Markowska-Kaczmar U., Mularczyk K.

Systems Science

|

2006

|

Vol. 32, no 4

35-43

EN

The paper describes a modification of the MulGex method in order to use it to extract rules from approximating neural network. Originally the method was designed to extract prepositional rules from classification neural network. The rules are searched by evolutionary algorithms working on two levels. The rules are optimized using the Pareto approach. The main principle referring to premise part of a rule has been unchanged but the form of conclusion instead of the class label describes a formula which can be a linear function and is encoded as a list of coefficients or it takes a form of a tree whose inner nodes contain functions and operators, and leaves - identifiers of attributes and numeric constants. Although the results obtained in the experiments for three different data sets can be assumed as satisfactory, some changes improving MulGex efficiency are proposed at the end.

7

Evolutionary approach to rule extraction from medical data

Kwaśnicka H., Markowska-Kaczmar U., Osojca T.

Journal of Medical Informatics & Technologies

|

2004

|

Vol. 7

KB3--12

EN

In the paper the method called CGA based on a cooperating genetic algorithm is presented. The CGA is developed for searching a set of rules describing classes in classification problems on the basis of training examples. The details of the method, such as a schema of coding (a chromosome), and a fitness function are shortly described. The method is independent of the type of attributes and it allows choosing different evaluation functions. Developed method was tested using different benchmark data sets. Next, in order to evaluate the efficiency of CGA, it was tested using the Breast Cancer data set with 10 fold cross validation technique.

8

Effect of fuzzy discretization in fuzzy rule-based systems for classification problems with continuous attributes

Ishibuchi H., Yamamoto T.

Archives of Control Sciences

|

2002

|

Vol. 12, no. 4

351-377

EN

Continuous attributes are usually discretized into intervals in machine learning and data mining. Our knowledge representation is, however, not always based on such discretization. For example, we usually use linguistic terms for dividing our ages into some categories with fuzzy boundaries. In this paper, we examine the effect of fuzzy discretization on the classification performance of fuzzy rule-based systems through computer simulations on simple numerical examples and real-world pattern classification problems. For exulting such computer simulations, we introduce a control parameter that specifies the overlap grade between adjacent antecedent fuzzy sets in fuzzy discretization. Interval discretization can be viewed as a special case of fuzzy discretization with no overlap. Computer simulations are performed using fuzzy discretization with various specifications of the overlap grade. Simulation results show that fuzzy rules have high generalization ability even when the domain interval of each continuous attribute is homogeneously partitioned into linguistic terms. On the other hand, generalization ability of rule-based systems strongly depends on the choice of theshold values in the case of interval discretization.

9

Effective Procedures for Handling Possible Equivalence Relations in Non-deterministic Information Systems

Sakai H.

Fundamenta Informaticae

|

2001

|

Vol. 48, nr 4

343-362

EN

This paper presents a tool for handling equivalence relations in non-deterministic information systems. Some applications of equivalence relations are also shown. In a deterministic information system, it is possible to define an equivalence relation for any set of attributes. However, in a non-deterministic information system, some kinds of equivalence relations which we call possible equivalence relations are definable. This paper proposes two effective procedures producing all possible equivalence relations for any non-deterministic information system and any set of attributes. The details of algorithms, the implementation of algorithms and applications to the rule extraction, etc. are presented.