Roman Neruda - Academia.edu (original) (raw)

Papers by Roman Neruda

Role-based frameworks enrich multi-agent system models with new organizational concepts: roles an... more Role-based frameworks enrich multi-agent system models with new organizational concepts: roles and groups. We present a formalization of a role-based approach in description logics serving as the common organizational model both for the development process and the management of the system. A central authority of role ontology agent is proposed in this paper. This agent performs management of the system: keeping track of the actual state of the system, querying the model and checking its social norms. Agents in the system can thus act in a sociable way especially in complicated configurations. Algorithms controlling agents in a computational intelligence modeling scenario are shown.

Due to many successful practical applications, deep neural networks and convolutional networks ha... more Due to many successful practical applications, deep neural networks and convolutional networks have become the state-of-art machine learning methods recently. The choice of network architecture for the task at hand is typically made by trial and error. This work deals with an automatic data-dependent architecture design. We propose an algorithm for optimization of architecture of convolutional network based on asynchronous evolution. The algorithm is inspired by and designed directly for the Keras library which is one of the most common implementations of deep neural networks. The proposed algorithm is successfully tested on MNIST and Fashion-MNIST data sets.

IEEE/WIC/ACM International Conference on Intelligent Agent Technology

We study and design multi-agent systems for computational intelligence modeling. Agents typically... more We study and design multi-agent systems for computational intelligence modeling. Agents typically reside in a high-performance parallel environment, such as a cluster of computational nodes, and utilize a non-blocking asynchronous communication. The need of accurate predictions of run-time and other characterizations of complex parallel asynchronous processes bring us to design a new parallel model creation methodology. In this article our approach is briefly described and a test case is shown and discussed.

Advances in Intelligent Systems and Computing, 2014

ABSTRACT This work describes a method to control a behaviour of intelligent data mining agent. We... more ABSTRACT This work describes a method to control a behaviour of intelligent data mining agent. We developed an adaptive decision making system that utilizes genetic programming technique to evolve an agent's decision strategy. The parameters of data mining task and current state of an agent are taken into account by tree structures evolved by genetic programming. Efficiency of decision strategies is compared from the perspectives of single and multi criteria optimization.

The robot localization problem is a fundamental and well studied problem in robotics research. Al... more The robot localization problem is a fundamental and well studied problem in robotics research. Algorithms used to estimate pose on the map are usually based on Kalman or particle filters. These algorithms are able to cope with errors, that arise due to inaccuracy of robot sensors and effectors. The performance of the localization algorithm depends heavily on their quality. This work shows performance of localization algorithm based on particle filter with small miniature low-cost E-puck robot. Information from VGA camera and eight infrared sensors are used to correct estimation of the robot's pose.

Lecture Notes in Computer Science, 2012

Lecture Notes in Computer Science, 2001

We present a learning algorithm for feedforward neural networks that is based on Kolmogorov theor... more We present a learning algorithm for feedforward neural networks that is based on Kolmogorov theorem concerning composition of ndimensional continuous function from one-dimensional continuous functions. A thorough analysis of the algorithm time complexity is presented together with serial and parallel implementation examples.

Artificial Neural Nets and Genetic Algorithms, 2003

A general concept of representation of connected groups of agents (schemes) within a multi-agent ... more A general concept of representation of connected groups of agents (schemes) within a multi-agent system is introduced and utilized for automatic building of schemes to solve a given computational task. We propose a combination of an evolutionary algorithm and a formal logic resolution system which is able to propose and verify new schemes. The approach is illustrated on simple examples.

Concepts, Opportunities and Challenges

In this chapter we address the problem of interoperability of incompatible business processes con... more In this chapter we address the problem of interoperability of incompatible business processes consisting of semantically annotated Web services. We propose a (semi-) automated solution, called process mediation, that analyzes the potentially incompatible process models of service requesters and service providers, identifies all incompatibilities, and automatically synthesizes reconciliation plans that can be used at runtime for resolving the identified incompatibilities. Compared to the previous research, the main novelty of our work is its strong focus on dynamic scenarios common for process-based service-oriented systems operating in open or semi-open dynamic environments. The chapter provides an overview of our recent body of work that covers techniques for automated incompatibilities detection and resolution, service discovery in the process mediation context, techniques for monitoring, and techniques for fault handling and recovery. We discuss the lessons we have learned and we also provide an outlook for future research directions.

2008 Sixth European Conference on Web Services, 2008

In most of research, service composition and discovery are treated separately. In the composition... more In most of research, service composition and discovery are treated separately. In the composition, it is assumed that primitive services are known to the composition component in advance. At the same time, in matchmaking algorithms, usually only one service is considered as a suitable candidate satisfying a query while service combinations are not allowed. The problem is that in realistic composition or mediation scenarios in dynamic environments some specific gap can be bridged only by using a combination of several services which are not known and need to be discovered. We consider such a situation and we propose an approach using a combined match and a composed match. We give a characterization of these two matching types and we develop appropriate matching algorithms.

Lecture Notes in Computer Science

Boolean matrix factorization is a generally accepted approach used in data analysis to explain da... more Boolean matrix factorization is a generally accepted approach used in data analysis to explain data or for data preprocessing in the supervised settings. In this paper we study factors in the supervised settings. We provide an experimental proof that factors are able to explain not only data as a whole but also classes in the data.

In this paper we compare the evolution of simple behaviour patterns for both an individual and a ... more In this paper we compare the evolution of simple behaviour patterns for both an individual and a group of simulated physical robots. An evolutionary algorithm with quite general objective function is used to study the ability to develop behaviour patterns such as the maze exploring ability. The group experiments demonstrate the development of collective behaviour where the group members follow the leader who is exploring the maze. Although controlled by identical simple neural network, the group members demonstrate a level of specialization. The experiments have been verified with the real physical Khepera robots.

We present various learning methods for RBF networks. The standard gradient-based learning is aug... more We present various learning methods for RBF networks. The standard gradient-based learning is augmented by the weighted norm adaptation. The three-step learning algorithm uses different unsupervised learning algorithms for setting the centroids. Two possible combinations with genetic learning algorithm are considered as well. All learning variants are thoroughly compared on two benchmark tasks.

A formalism for the logical description of computational agents and multi-agent systems is given.... more A formalism for the logical description of computational agents and multi-agent systems is given. It is explained how it such a formal description can be used to configure and reason about multi-agent systems realizing computational intelligence models. A usage within a real software system Bang 3 is demonstrated. The logical description of multi- agent systems opens Bang 3 for interaction with ontology based distributed knowledge systems like the Semantic Web or the Knowledge Grid.

PCT No. PCT/DK86/00011 Sec. 371 Date Nov. 26, 1986 Sec. 102(e) Date Nov. 26, 1986 PCT Filed Jan. ... more PCT No. PCT/DK86/00011 Sec. 371 Date Nov. 26, 1986 Sec. 102(e) Date Nov. 26, 1986 PCT Filed Jan. 31, 1986 PCT Pub. No. WO86/04538 PCT Pub. Date Aug. 14, 1986.Apparatus for cutting pieces of a predetermined length from a string of material by means of a rotary saw blade, said string being conveyed intermittently and at irregular speed. To meet this object without the apparatus being more than modestly space-demanding at the three sides of the string, the saw blade is mounted on a transmission (9), which is mounted on a lever (4) extending substantially at a right angle to the string substantially where the cutting is to be performed and pivotable around an axle (5) perpendicular to the string, which lever comprises gripping means (13, 14) for holding the lever unmovable with respect to the string, the transmission (9) being displaceable with respect to the lever to provide the feeding of the saw blade. The apparatus is used for cutting extruded strings of frozen food, an apparatus being placed at each of several strings being extruded simultaneously at an acute angle with respect to the axis of the extruder.

International Journal of Agent-Oriented Software Engineering, 2009

The ability to deal with the incompatibilities of service requesters and providers is a critical ... more The ability to deal with the incompatibilities of service requesters and providers is a critical factor for achieving interoperability in dynamic open environments. We focus on the problem of process mediation of the semantically annotated process models of the service requester and service provider. We propose an Abstract Process Mediation Framework (APMF) identifying the key functional areas that need to be addressed by process mediation components. Next, we present algorithms for solving the process mediation problem in two scenarios: (1) when the mediation process has complete visibility of the process model of the service provider and service requester (complete visibility scenario) and (2) when the mediation process has visibility only of the process model of the service provider, but not the service requester (asymmetric scenario). The algorithms combine planning and semantic reasoning with the discovery of appropriate external services such as data mediators. Finally, the Process Mediation Agent (PMA) is introduced, which realises an execution infrastructure for runtime mediation.

Usage of genetic algorithms in the Information retrieval area, especially in optimizing a Boolean... more Usage of genetic algorithms in the Information retrieval area, especially in optimizing a Boolean query, is presented in this paper. The proposed evolution of Boolean queries should increase the performance of the information retrieval system by means of the precision and recall. Influence of both criteria is thorougly tested for different genetic algorithm settings. The quality improvement achieved by our approach is discussed.

Proceedings of the 2017 Federated Conference on Computer Science and Information Systems, 2017

Deep neural networks enjoy high interest and have become the state-of-art methods in many fields ... more Deep neural networks enjoy high interest and have become the state-of-art methods in many fields of machine learning recently. Still, there is no easy way for a choice of network architecture. However, the choice of architecture can significantly influence the network performance. This work is the first step towards an automatic architecture design. We propose a genetic algorithm for an optimization of a network architecture. The algorithm is inspired by and designed directly for the Keras library [1] that is one of the most common implementations of deep neural networks. The target application is the prediction of air pollution based on sensor measurements. The proposed algorithm is evaluated on experiments on sensor data and compared to several fixed architectures and support vector regression.