Találati lista | Tudóstér

001-es BibID:	BIBFORM062529
035-os BibID:	Zbl.06476013
Első szerző:	Gál Zoltán (informatikus)
Cím:	Surprise event detection of the supercomputer execution queues / Zoltán Gál, Tibor Tajti, György Terdik
Dátum:	2015
ISSN:	1787-5021 1878-6117
Megjegyzések:	Huge amount of data is generated by and collected from the IoT (Internet of Things) physical and virtual devices. These sets of data series reflect in complex form the state of a given system in multidimensional space. Healthiness evaluation of a given system implies state analysis with enhanced methods. Special events can appear during the execution of jobs in a supercomputer (HPC - High Performance Computer) system. Depending on the HPC architecture hundreds or thousands of computation nodes are working in parallel. The scheduler of the HPC front-end node manages different queues (parallel, serial, test, etc.) of the job execution. The multitude of data series captured periodically with several tens of thousands of samples creates a set of several dozen variables for each computation node. The healthiness of the whole HPC system is a temporal concept in the term of 2D or 4D multidimensional time-space domains. In this paper we propose a healthiness evaluation method for each execution queue of a two different HPC system with 20 TFLOP/s and 5 TFLOP/s computation capacities, respectively. Time independent community structure is determined and controlled based on multiple similarity measures and ANN (Artificial Neural Network) based SOM (Self-Organized Map) algorithm. For each cluster of variables is determined a representing variable, including time specific and global characteristics of the own cluster. The resulting set of representing variables contains less than ten dissimilar time series. Wavelet methods are used for extreme event detection in time of each representing variable. The surprise event detection in time of the HPC execution queues is based on the simultaneity of extreme event fingerprints.
Tárgyszavak:	Műszaki tudományok Informatikai tudományok idegen nyelvű folyóiratközlemény hazai lapban High Performance Computer sensors/actuators IoT Complex Event Processing Event Stream Processing Artificial Neural Networks SOM FFT STFT Wavelets
Megjelenés:	Annales Mathematicae et Informaticae. - 44 (2015), p. 87-97. -
További szerzők:	Tajti Tibor Gábor (1970-) (informatikus) Terdik György (1949-) (matematikus, informatikus)
Pályázati támogatás:	TÁMOP-4.2.2.C-11/1/KONV-2012-0010 TÁMOP
Internet cím:	Szerző által megadott URL Intézményi repozitóriumban (DEA) tárolt változat
Borító:
Saját polcon:

001-es BibID:	BIBFORM088788
035-os BibID:	(WoS)000600058700018 (Scopus)85097908666
Első szerző:	Tajti Tibor Gábor (informatikus)
Cím:	New voting functions for neural network algorithms / Tibor Tajti
Dátum:	2020
ISSN:	1787-5021 1787-6117
Megjegyzések:	Neural Network and Convolutional Neural Network algorithms are among the best performing machine learning algorithms. However, the performance of the algorithms may vary between multiple runs because of the stochastic nature of these algorithms. This stochastic behavior can result in weaker accuracy for a single run, and in many cases, it is hard to tell whether we should repeat the learning giving a chance to have a better result. Among the useful techniques to solve this problem, we can use the committee machine and the ensemble methods, which in many cases give better than average or even better than the best individual result. We defined new voting function variants for ensemble learner committee machine algorithms which can be used as competitors of the well-known voting functions. Some belong to the locally weighted average voting functions, others are meta voting functions calculated from the output of the previous voting functions functions called with the results of the individual learners. The performance evaluation of these methods was done from numerous learning sessions.
Tárgyszavak:	Természettudományok Matematika- és számítástudományok idegen nyelvű folyóiratközlemény hazai lapban folyóiratcikk Machine learning neural networks committee machines ensemble methods
Megjelenés:	Annales Mathematicae et Informaticae. - 52 (2020), p. 229-242. -
Internet cím:	Szerző által megadott URL DOI Intézményi repozitóriumban (DEA) tárolt változat
Borító:
Saját polcon:

001-es BibID:	BIBFORM088787
035-os BibID:	(WoS)000600058700017 (Scopus)85098076385
Első szerző:	Tajti Tibor Gábor (informatikus)
Cím:	Fuzzification of training data class membership binary values for neural network algorithms / Tibor Tajti
Dátum:	2020
ISSN:	1787-5021 1787-6117
Megjegyzések:	We propose an algorithm improvement for classifying machine learning algorithms with the fuzzification of training data binary class membership values. This method can possibly be used to correct the training data out- put values during the training. The proposed modification can be used for algorithms running individual learners and also as an ensemble method for multiple learners for better performance. For this purpose, we define the single and the ensemble variants of the algorithm. Our experiment was done using convolutional neural network (CNN) classifiers for the base of our pro- posed method, however, these techniques might be used for other machine learning classifiers as well, which produce fuzzy output values. This fuzzi- fication starts with using the original binary class membership values given in the dataset. During training these values are modified with the current knowledge of the machine learning algorithm.
Tárgyszavak:	Természettudományok Matematika- és számítástudományok idegen nyelvű folyóiratközlemény hazai lapban folyóiratcikk machine learning neural networks fuzzification
Megjelenés:	Annales Mathematicae et Informaticae. - 52 (2020), p. 217-228. -
Internet cím:	Szerző által megadott URL DOI Intézményi repozitóriumban (DEA) tárolt változat
Borító:
Saját polcon: