Wyniki wyszukiwania - Biblioteka Nauki

1

Monte Carlo Tree Search Algorithm for the Euclidean Steiner Tree Problem

100%

Bereta M.

Journal of Telecommunications and Information Technology

|

2017

|

tom nr 4

71--81

EN

This study is concerned with a novel Monte Carlo Tree Search algorithm for the problem of minimal Euclidean Steiner tree on a plane. Given p p p points (terminals) on a plane, the goal is to ﬁnd a connection between all the points, so that the total sum of the lengths of edges is as low as possible, while an addition of extra points (Steiner points) is allowed. Finding the minimum Steiner tree is known to be np-hard. While exact algorithms exist for this problem in 2D, their eﬃciency decreases when the number of terminals grows. A novel algorithm based on Upper Conﬁdence Bound for Trees is proposed. It is adapted to the speciﬁc characteristics of Steiner trees. A simple heuristic for fast generation of feasible solutions based on Fermat points is proposed together with a correction procedure. By combing Monte Carlo Tree Search and the proposed heuristics, the proposed algorithm is shown to work better than both the greedy heuristic and pure Monte Carlo simulations. Results of numerical experiments for randomly generated and benchmark library problems (from OR-Lib) are presented and discussed.

2

Entropy-based regularization of AdaBoost

100%

Bereta M.

Computer Assisted Methods in Engineering and Science

|

2017

|

tom Vol. 24, no. 2

89--100

EN

In this study, we introduce an entropy-based method to regularize the AdaBoost algorithm. The AdaBoost algorithm is a well-known algorithm used to create aggregated classifiers. In many real-world classification problems in addition to paying special attention classification accuracy of the final classifier, great focus is placed on tuning the number of the so-called weak learners, which are aggregated by the final (strong)classifier. The proposed method is able to improve the AdaBoost algorithm in terms of both criteria. While many approaches to the regularization of boosting algorithms can be complicated, the proposed method is straightforward and easy to implement. We compare the results of the proposed method (EntropyAdaBoost) with the original AdaBoost and also with its regularized version, є-AdaBoost on several classification problems. It is shown that the proposed methods of EntropyAdaBoost and є-AdaBoost are strongly complementary when the improvement of AdaBoost is considered.

3

Combining negative selection with immune K-means algorithm for improving the support vector machines method

63%

Bereta M. , Burczyński T.

Prace Naukowe Politechniki Warszawskiej. Elektronika

|

2007

|

tom z. 160

25-33

EN

This paper presents a novel method of using the ideas from Artificial Immune Systems for improving the performance of the support Vector Machines. By means of Immune K-Means algorithm a set of artificial data is generated based on the oryginal training data. The artificial data describes the most important information from the classifiers learning point of view - the information about the boundaries among the classes remain in the artificial data. Combining the Immune K-Means algorithm with Negative Selection methods allows for further improvements of the artificial data set. The proposed approach allows to speed up the learning process of SVM when the training data set is large by extracting the most important information first. The proposed method can also be used as a data compression, especially suited when the information about boundaries among classes is an important issue. The artificial data can be created once and then used for parameters tuning of different classification methods, speeding up the learning process.

4

The use of urban indicators in forecasting a real estate value with the use of deep neural network

63%

Bazan-Krzywoszańska A. , Bereta M.

Reports on Geodesy and Geoinformatics

|

2018

|

tom Vol. 106

25--34

EN

Records of municipal planning documents directly affect the land use. In this way, the market price of the land is also shaped. Awareness of the economic and social consequences of adapting specific solutions is the primary argument that should condition the local policy in terms of spatial planning. The research results indicate that the network trained with attributes which do not describe a property value by its price was able to estimate it with acceptable and satisfactory results. The possibility to use artificial multilayer networks in spatial policy decision-making seems well founded. The research results show the relevance of the assumption that using them for modeling can be helpful in selecting the most advantageous variant of planning arrangements in a local law document which determines the land use and development, therefore impacts its value.

5

Evolving ensembles of linear classifiers by means of clonal selection algorithm

63%

Bereta M. , Burczyński T.

Control and Cybernetics

|

2010

|

tom Vol. 39, no 2

325-342

EN

Artificial immune systems (AIS) have become popular among researchers and have been applied to a variety of tasks. Developing supervised learning algorithms based on metaphors from the immune system is still an area in which there is much to explore. In this paper a novel supervised immune algorithm based on clonal selection framework is proposed. It evolves a population of linear classifiers used to construct a set of classification rules. Aggregating strategies, such as bagging and boosting, are shown to work well with the proposed algorithm as the base classifier.

6

Immune K-Means : a novel immune algorithm for data clustering and multiple-class discrimination

63%

Bereta M. , Burczyński T.

Prace Naukowe Politechniki Warszawskiej. Elektronika

|

2006

|

tom z. 156

49-60

EN

This paper presents a novel approach to data clustering and multiple-class classification problems. The proposed method is based on a metaphor derived from immune systems, the clonal selection paradigm. A novel clonal selection algorithm - Immune K-Means, is proposed. The proposed system is able to cluster real valued data efficiently and correctly, dynamically estimating the number of clusters. In classification problems discrimination among classes is based on the k-nearest neighbor method. Two different types of suppression are proposed. They enable the evolution of different populations of lymphocytes well suited to a given problem : clustering or classification. The first type of suppression enables the lymphocytes to discover the data distribution while the second type of suppression focuses the lymphocytes on the classes' boundaries. Primary results on artificial data and a real-world benchmark dataset (Fisher's Iris Database) as well as a discussion of the parameters of the algorithm are given.

7

Opracowanie eukariotycznego systemu ekspresyjnego dzialajacego na podstawie aktywnosci rekombinazy Flp

38%

Wicher K. B. , Jura N. , Kowanetz M. , Lorenowicz M. , Rafalska I. , Szybalski W. , Bereta M.

Biotechnologia

|

1999

|

nr 1

144-152

8

Pulsating electromagnetic field stimulation prevents cell death of puromycin treated U937 cell line

32%

Kaszuba-Zwoinska J. , Wojcik K. , Bereta M. , Ziomber A. , Pierzchalski P. , Rokita E. , Marcinkiewicz J. , Zaraska W. , Thor P.

Journal of Physiology and Pharmacology

|

2010

|

tom 61

|

nr 2

EN

Aim of study was to verify whether pulsating electromagnetic field (PEMF) can affect cancer cells proliferation and death. U937 human lymphoid cell line at densities starting from 1x106 cells/ml to 0.0625x106 cells/ml, were exposed to a pulsating magnetic field 50Hz, 45±5 mT three times for 3 h per each stimulation with 24 h intervals. Proliferation has been studied by counting number of cells stimulated and non-stimulated by PEMF during four days of cultivation. viability of cells was analyzed by APC labeled Annexin V and 7-AAD (7-amino-actinomycin D) dye binding and flow cytometry. Growing densities of cells increase cell death in cultures of U937 cells. PEMF exposition decreased amount of cells only in higher densities. Measurement of Annexin V binding and 7-AAD dye incorporation has shown that density-induced cell death corresponds with decrease of proliferation activity. PEMF potentiated density-induced death both apoptosis and necrosis. The strongest influence of PEMF has been found for 1x106cells/ml and 0.5x106 cells/ml density. To eliminate density effect on cell death, for further studies density 0.25x106 cells/ml was chosen. Puromycin, a telomerase inhibitor, was used as a cell death inducer at concentration 100 µg/ml. Combined interaction of three doses of puromycin and three fold PEMF interaction resulted in a reduced of apoptosis by 24,7% and necrosis by 13%. PEMF protects U937 cells against puromycin- induced cell death. PEMF effects on the human lymphoid cell line depends upon cell density. Increased density induced cells death and on the other hand prevented cells death induced by puromycin.