Wyniki wyszukiwania - Biblioteka Nauki

1

Near Sets. Special Theory about Nearness of Objects

100%

Peters J. F.

|

2007

|

tom Vol. 75, nr 1-4

407-433

EN

The problem considered in this paper is how to approximate sets of objects that are qualitatively but not necessarily spatially near each other. The term qualitatively near is used here to mean closeness of descriptions or distinctive characteristics of objects. The solution to this problem is inspired by the work of Zdzisaw Pawlak during the early 1980s on the classification of objects by means of their attributes. This article introduces a special theory of the nearness of objects that are either static (do not change) or dynamic (change over time). The basic approach is to consider a link relation, which is defined relative to measurements associated with features shared by objects independent of their spatial relations. One of the outcomes of this work is the introduction of new forms of approximations of objects and sets of objects. The nearness of objects can be approximated using rough set methods. The proposed approach to approximation of objects is a straightforward extension of the rough set approach to approximating objects, where approximation can be considered in the context of information granules (neighborhoods). In addition, the usual rough set approach to concept approximation has been enriched by an increase in the number of granules (neighborhoods) associated with the classification of a concept as near to its approximation. A byproduct of the proposed approximation method is what we call a near set. It should also be observed that what is presented in this paper is considered a special (not a general) theory about nearness of objects. The contribution of this article is an approach to nearness as a vague concept which can be approximated from the state of objects and domain knowledge.

2

Reinforcement Learning with Approximation Spaces

63%

Peters J. F. , Henry C.

|

2006

|

tom Vol. 71, nr 2,3

323-349

EN

This paper introduces a rough set approach to reinforcement learning by swarms of cooperating agents. The problem considered in this paper is how to guide reinforcement learning based on knowledge of acceptable behavior patterns. This is made possible by considering behavior patterns of swarms in the context of approximation spaces. Rough set theory introduced by Zdzisaw Pawlak in the early 1980s provides a ground for deriving pattern-based rewards within approximation spaces. Both conventional and approximation space-based forms of reinforcement comparison and the actor-critic method as well as two forms of the off-policy Monte Carlo learning control method are investigated in this article. The study of swarm behavior by collections of biologically-inspired bots is carried out in the context of an artificial ecosystem testbed. This ecosystem has an ethological basis that makes it possible to observe and explain the behavior of biological organisms that carries over into the study of reinforcement learning by interacting robotic devices. The results of ecosystem experiments with six forms of reinforcement learning are given. The contribution of this article is the presentation of several viable alternatives to conventional reinforcement learning methods defined in the context of approximation spaces.

3

A Comparision of Different Decision Algorithms Used in Volumetric Storm Cells Classification

51%

Suraj Z. , Peters J. F. , Rząsa W.

|

2002

|

tom Vol. 51, nr 1,2

201-214

EN

Decision algorithms useful in classifying meteorological volumetric radar data are the subject of described in the paper experiments. Such data come from the Radar Decision Support System (RDSS) database of Environment Canada and concern summer storms created in this country. Some research groups used the data completed by RDSS for verifying the utility of chosen methods in volumetric storm cells classification. The paper consists of a review of experiments that were made on the data from RDSS database of Environment Canada and presents the quality of particular classifiers. The classification accuracy coefficient is used to express the quality. For five research groups that led their experiments in a similar way it was possible to compare received outputs. Experiments showed that the Support Vector Machine (SVM) method and rough set algorithms which use object oriented reducts for rule generation to classify volumetric storm data perform better than other classifiers.

4

Towards an Ontology of Approximate Reason

51%

Peters J. F. , Skowron A. , Stepaniuk J. , Ramanna S.

|

2002

|

tom Vol. 51, nr 1,2

157-173

EN

This article introduces structural aspects in an ontology of approximate reason. The basic assumption in this ontology is that approximate reason is a capability of an agent. Agents are designed to classify information granules derived from sensors that respond to stimuli in the environment of an agent or received from other agents. Classification of information granules is carried out in the context of parameterized approximation spaces and a calculus of granules. Judgment in agents is a faculty of thinking about (classifying) the particular relative to decision rules derived from data. Judgment in agents is reflective, but not in the classical philosophical sense (e.g., the notion of judgment in Kant). In an agent, a reflective judgment itself is an assertion that a particular decision rule derived from data is applicable to an object (input). That is, a reflective judgment by an agent is an assertion that a particular vector of attribute (sensor) values matches to some degree the conditions for a particular rule. In effect, this form of judgment is an assertion that a vector of sensor values reflects a known property of data expressed by a decision rule. Since the reasoning underlying a reflective judgment is inductive and surjective (not based on a priori conditions or universals), this form of judgment is reflective, but not in the sense of Kant. Unlike Kant, a reflective judgment is surjective in the sense that it maps experimental attribute values onto the most closely matching descriptors (conditions) in a derived rule. Again, unlike Kant's notion of judgment, a reflective judgment is not the result of searching for a universal that pertains to a particular set of values of descriptors. Rather, a reflective judgment by an agent is a form of recognition that a particular vector of sensor values pertains to a particular rule in some degree. This recognition takes the form of an assertion that a particular descriptor vector is associated with a particular decision rule. These considerations can be repeated for other forms of classifiers besides those defined by decision rules.

5

Approaches to Conflict Dynamics Based on Rough Sets

51%

Ramanna S. , Peters J. F. , Skowron A.

|

2007

|

tom Vol. 75, nr 1-4

453-468

EN

Conflict analysis and conflict resolution play an important role in negotiation during contract-management situations in many organizations. The issue here is how to model a combination of complex situations among agents where there are disagreements leading to a conflict situation, and there is a need for an acceptable set of agreements. Conflict situations also result due to different sets of view points about issues under negotiation. The solution to this problem stems from pioneering work on this subject by Zdzisaw Pawlak, which provides a basis for a complex conflict model encapsulating a decision system with complex decisions. Several approaches to the analysis of conflicts situations are presented in this paper, namely, conflict graphs, approximation spaces and risk patterns. An illustrative example of a requirements scope negotiation for an automated lighting system is presented. The contribution of this paper is a rough set-based requirements scope determination model and assessment mechanisms using a complex conflict model.

6

An application of rough set methods in control design

51%

Peters J. F. , Skowron A. , Suraj Z.

|

2000

|

tom Vol. 43, Nr 1-4

269-290

EN

The paper deals with an automatic concurrent control design method derived from the specification of a discrete event control system represented in the form of a decision table. The main stages of our approach are: the control specification by decision tables, generation of rules from the specification of the system behavior, and converting rules set into a concurrent program represented in the form of a Petri net. Our approach is based on rough set theory.

7

Rough C-means and Fuzzy Rough C-means for Colour Quantisation

45%

Schaefer G. , Hu Q. , Zhou H. , Peters J. F. , Hassanien A. E.

|

2012

|

tom Vol. 119, nr 1

113-120

EN

Colour quantisation algorithms are essential for displaying true colour images using a limited palette of distinct colours. The choice of a good colour palette is crucial as it directly deter- mines the quality of the resulting image. Colour quantisation can also be seen as a clustering problem where the task is to identify those clusters that best represent the colours in an image. In this paper we propose rough c-means and fuzzy rough c-means clustering algorithms for colour quantisation of images. Both approaches utilise the concept of lower and upper approximations of clusters to define palette colours. While in the rough c-means approach cluster centroids are refined iteratively through a linear combination of elements of the lower and upper approximations, the fuzzy rough c-means technique assigns variable membership values to the elements in the boundary region which in turn are incorporated into the calculation of cluster centres. Experimental results on a standard set of images show that these approaches performs significantly better than other, purpose built colour quantisation algorithms.

8

Sensor, Filter, and Fusion Models with Rough Petri Nets

45%

Peters J. F. , Ramanna S. , Borkowski M. , Skowron A. , Suraj Z.

|

2001

|

tom Vol. 47, nr 3,4

307-323

EN

This paper considers models of sensors, filters, and sensor fusion with Petri nets defined in the context of rough sets. Sensors and filters are fundamental computational units in the design of systems. The intent of this work is to construct Petri nets to simulate conditional computation in approximate reasoning systems, which are dependent on filtered input from selected sensors considered relevant in problem solving. In this paper, coloured Petri nets provide a computational framework for the definition of a family of Petri nets based on rough set theory. Sensors are modeled with what are known as receptor processes in rough Petri nets. Filters are modeled as ukasiewicz guards on some transitions in rough Petri nets. A ukasiewicz guard is defined in the context of multivalued logic. ukasiewicz guards are useful in culling from a collection of active sensors those sensors with the greatest relevance in a problem-solving effort such as classification of a "perceived" phenomenon in the environment of an agent. The relevance of a sensor is computed using a discrete rough integral. The form of sensor fusion considered in this paper consists in selecting only those sensors considered relevant in solving a problem. The contribution of this paper is the modeling of sensors, filters, and fusion in the context of receptor processes, ukasiewicz guards, and rough integration, respectively.