Rudolf JAKSA
@matsuko.com
MATSUKO
RESEARCH, TEACHING, or OTHER INTERESTS
Artificial Intelligence
Scopus Publications
Scholar Citations
Scholar h-index
Scholar i10-index
Scopus Publications
Peter Polak, Rudolf Jaksa, and Jan Vascak
IEEE
This paper is focused on the application of fractal analysis in the attention management of humanoid robot. We designed a fuzzy controller to combine the face detection, movement detection and the fractal dimension signals to control the head movement of robot Nao. Also, the gaze problem is addressed by the controller. Implementation details are included in the paper, including configuration parameters, which we found optimal according to subjective analysis and possibilities of current hardware. We found the fuzzy controller to be advantageous for implementation of attention manager because of smoothing of the movement of robot when compared to the simple rule based implementation, and also because the fuzzy controller implementation of manager is more clear than a naive if-then heuristics code. We also found the fractal dimension to be useful additional signal for attention management of robot, which can be computed in near real-time on current hardware and static input images.
Rudolf Jakša, Martina Zeleňáková, Juraj Koščák, and Helena Hlavatá
VGTU Technika
The paper is focused on analysis of local neural network model of precipitation. We use basic multilayer perceptron neural network with the time-window on input data to predict the precipitation. We predict the precipitation in the next day from the local meteorological data from past days. Data from the past 60 years were used to train the predictor. Obtained prediction model is specific for given area of Košice City in Slovakia, as the prediction is based on the statistics of the weather in given area. This precipitation predictor is multiple-input-single-output architecture with a single value per day resolution on output. Obtained results show that good local temperature prediction accuracy is possible with chosen setup, but it is worse for the precipitation prediction. Also the training requirements of precipitation predictor seem to be significantly higher then for the temperature predictor. Obtained prediction results can be used for applications based on local meteorological station data, although they are not as accurate as the state of art agency predictions based on satellite data. In the paper we will analyze design of the precipitation predictor based on existing design of the temperature predictor and provide the reader with recommended setup of such predictor for application with his/her local precipitation data.
Denis Vere, Rudolf Jaka, and Jakub Hvizdo
IEEE
This paper is focused on the problem of tracking an object by the head movement of robot with two cameras simultaneously, one robot camera and one fixed external camera. The goal of using external camera is to test, how it can aid the head camera of robot when the object moves out of its field of view. The setup of system is focused on comparison of robot with and without additional camera. The tracked object is a simple pink ball and the tracking mechanism is a simple multilayer perceptron with the backpropagation algorithm. The mechanism of tracking is not to shift the coordinates of moving object on fixed scene, but actually to turn the head of robot to have the object in the center of scene, like with a human tracking some object. In the paper we provide details of setup of working system, with results and parameters, which can be used as a starting point for similar experiments. Actually, the training set construction is important for this type of robotic tracking, please see details in the paper. The system is working. We believe, it is easy to repeat our results, and the effect of added camera can be demonstrated.
Rastislav Rusnak and Rudolf Jaksa
IEEE
This paper deals with stochastic weight update methods for neural networks learning. We will study two methods, stochastic weights selection and stochastic neurons selection. These methods have to allow better parallelization of the backpropagation algorithm, although in this paper we will use only the conventional serial implementation. We will use meteorological data for experimentation with neural networks based weather prediction. We will show that proposed methods can be used to replace regular backpropagation, but in the serial implementation they are not efficient.
Ján Vaščák, Rudolf Jakša, Juraj Koščák, and Ján Adamčák
Elsevier BV
Juraj Koscak, Rudolf Jaksa, and Peter Sincak
IEEE
Stochastic weight update is a variant of error back-propagation algorithm for learning of artificial neural networks. It allows for efficient topology-independent implementation of backpropagation through time for recurrent networks. In stochastic weight update scenario, constant number of weights and neurons is randomly selected and updated. This is in contrast to the classical ordered update, where all weights/neurons are always updated. In this paper we will study performance of stochastic weight update on recurrent neural networks using concept of feedforward network with added recurrent neurons.
Ján Adamčák, Rudolf Jakša, and Ján Liguš
SCITEPRESS - Science and and Technology Publications
In this paper we describe how to build a fully autonomous system for collection, prediction and presentation of single-position meteorological data the local weather prediction system. By employing nonlinear statistics with neural network predictor on meteorological time-series data we were able to achieve good results for the one-day weather prediction. This novel local statistical approach to weather prediction is different compare to standard methods which are based on the air mass movement modelling. Main objective of this paper is to describe whole system for local weather prediction including technology, software, methods and parameters, and also experimental results.
Rudolf Jaksa
IEEE
This paper proposes the standard specification of metric numbers and is intended to discuss the topic. The numbers combined with metric prefixes like the 2k for the year, or 2k2 for resistors inscription are popularly used. However, there exists no standard to guide the use of such numbers, and typical precision is limited to few digits, as in the E24 series used for electronic parts. Computational methods rely heavily on the numerical data, as opposed to the rest of computer science or artificial intelligence, which are often built around textual data. To communicate numerical data, either with human, or between programs, several numbers representations are available. In this paper we want to show, that proposed metric numbers are a useful addition to the computational cybernetics paradigm. We will discuss and try to resolve problems related to the implementation of metric numbers system. Among these issues, we intend to standardize the suffix notation, as is used in the 2k2 resistor, and the plus/minus notation for sparse numbers, like in the 2M+1. We will also discuss and provide the reference implementation in the end.
Juraj Koščák, Rudolf Jakša, Rudolf Sepeši, and Peter Sinčák
Springer International Publishing
Juraj Koščák, Rudolf Jakša, and Peter Sinčák
Springer Berlin Heidelberg
P. Smolar, P. Sincak, and R. Jaksa
IEEE
This work is deals with processing and analysis of ECG waves, namely with recognition of ECG samples with diagnosis of myocardial infarct and arrhythmia from samples. As a base concept for comparing the ECG wave to the typical wave,Template matching method is used, which can find the best similarity between the test sample and ECG templates. With respect to the metrics it calculates their relative similarity, too. Input data were obtained from the project PhysioNet, gathered at the Institute of Cardiology at the University Clinic Benjamin Franklin in Berlin and digitalized in the National Metrology Institute, Germany under the name PTB ECG database. The outputs are the similarity coefficients of the twelve conventional ECG leads and the six basic parameters of waves. The results of our proposal with used methods for data preprocessing and implemented algorithm are comparable with the results obtained by systems based on neural networks classification. It has the potential to help physicians in the initial analysis and identification of the patient's condition.
Juraj Koscak, Rudolf Jaksa, and Peter Sincak
IEEE
We will examine stochastic weight update in the backpropagation algorithm on feed-forward neural networks. It was introduced by Salvetti and Wilamowski in 1994 in order to improve probability of convergence and speed of convergence. However, this update method has also one another quality, its implementation is simple for arbitrary network topology. In stochastic weight update scenario, constant number of weights is randomly selected and updated. This is in contrast to classical ordered update, where always all weights are updated. We will describe exact implementation, and present example results on toy-task data with feed-forward neural network topology. Stochastic weight update is suitable to replace classical ordered update without any penalty on implementation complexity and with good chance without penalty on quality of convergence.
Miron Kuzma, Rudolf Jaksa, and Peter Sincak
IEEE
Clustering of users inputs in multi-user Interactive Evolutionary Computation is intended to allow to collect large data sets for user-behavior modeling, while preserving the user's individuality. By clustering the user input data, we group together similar behaviors and distribute opposite ones, thus preventing conflicts in data resulting from opposite opinions of different users. Without this clustering, it might be not possible to use data obtained from several users for the behavior modeling. This paper tries to present the application of Self-Organizing-Map clustering in the task of font design with Interactive Evolutionary Computation interface.
Matúš Užák, Rudolf Jakša, and Peter Sinčák
Springer Berlin Heidelberg
Matus Uzak, Igor Vertal', Rudolf Jaksa, and Peter Sincak
IEEE
Visualization of the learning of neural network faces the problem of dealing with overwhelming amount of visual information. This paper describes the application of clustering methods for reduction of visual information in the response function visualization. When only clusters of neurons are visualized, instead of direct visualization of responses of all neurons in the network, the amount of visually presented information can be significantly reduced. This is useful for reducing user fatigue and also for minimizing the visualization equipment requirements. We show, that application of Kohonen network or growing neural gas with utility factor algorithm allows to visualize the learning of moderate-sized neural networks in real time. Comparison of both algorithms in this task is provided, also with performance analysis and example results of response function visualization.
Pavol Malinak and Rudolf Jaksa
IEEE
In this paper, two novel methods, BP/ES and ES/LMS, for simultaneous gradient and evolutionary adaptation of weights of neural network are proposed. In the BP/ES, an evolution strategy is used to optimize the last layer of the multilayer perceptron type of neural network, and back- propagation algorithm trains the rest of the network. The main idea of ES/LMS is to employ the least mean square algorithm to adapt the last layer of network and evolution strategy to optimize the rest of the network. Hybrid approaches to neural network learning, based on gradient and evolutionary techniques combinations, are aimed to raise the advantages of both approaches mentioned above - reliable computational requirements of gradient techniques and global search capabilities of evolutionary approaches. In general, neural network hybrid learning approaches are usually "sequential", rather than simultaneous. In the first step, the evolutionary technique is used to locate a promising region in the search space, and then the gradient technique is employed for fine tuning of network parameters in this region. The proposed BP/ES and ES/LMS methods investigate different approach. They perform "spatial" synthesis of gradient and evolutionary techniques, in which the neural network is partitioned into two parts - output layer versus the other layers - which are adapted simultaneously, however using these two different methods. Experimental results with error back-propagation algorithm, evolution strategies with and without covariances, BP/ES and ES/LMS method on the benchmark "XOR" and "circle in square" data are provided.
Matúš Užák and Rudolf Jakša
Springer Berlin Heidelberg