Informacja

Drogi użytkowniku, aplikacja do prawidłowego działania wymaga obsługi JavaScript. Proszę włącz obsługę JavaScript w Twojej przeglądarce.

English (EN)·Polski (PL)·Yкраїнський (UK)
Przejdź do strony domowej biblioteki Centralna Biblioteka Wojskowa Link otwiera się w nowym oknie Logo biblioteki Prolib Integro - strona głównaCentralna Biblioteka Wojskowa

Wyszukujesz frazę "artificial" wg kryterium: Temat

  Lista filtrów

Dostawca treści

Wyrównaj
Wyświetlanie 1-15 z 562
Tytuł:
The road to conscious machines: AI through failed ideas
Autorzy:
Krzanowski, Roman
Powiązania:
https://bibliotekanauki.pl/articles/1943050.pdf
Data publikacji:
2021-11-07
Wydawca:
Copernicus Center Press
Tematy:
artificial intelligence
history of artificial intelligence
artificial intelligence paradigms
future of artificial intelligence
Opis:
Presented here is a review of the book by Michael Wooldridge The Road to Conscious Machines: The Story of AI. The book was published in 2021 by Pelican Publishing Company. The book presents a philosophical exploration of the AI-related quest to understand the essence of our minds, ethics, and consciousness. Nonetheless, the book is unconventional as it focuses on failures of past AI programs rather than on AI success stories. The author also indicates the possible pathways to develop new, more efficient AI paradigms. The short technical section at the end of the book offers more in-depth information on topics like Prolog, Bayesian, and deep neural networks.
Źródło:
Zagadnienia Filozoficzne w Nauce; 2021, 70; 171-181
0867-8286
2451-0602
Pojawia się w:
Zagadnienia Filozoficzne w Nauce
Dostawca treści:
Biblioteka Nauki
Artykuł
Tytuł:
Jeszcze raz o ucieleśnionej sztucznej inteligencji
Embodied artificial intelligence once again
Autorzy:
Sarosiek, Anna
Powiązania:
https://bibliotekanauki.pl/articles/690932.pdf
Data publikacji:
2017
Wydawca:
Copernicus Center Press
Tematy:
artificial intelligence
good old-fashioned artificial intelligence
embodied artificial intelligence
embodied mind
Opis:
Book review of: Vincent C. Müller (ed.), Fundamental issues of artificial intelligence, Springer International Publishing AG, 2016, pp. 570.
Recenzja książki: Vincent C. Müller (red.), Fundamental issues of artificial intelligence, Springer International Publishing AG, 2016, ss. 570.
Źródło:
Zagadnienia Filozoficzne w Nauce; 2017, 63; 231-240
0867-8286
2451-0602
Pojawia się w:
Zagadnienia Filozoficzne w Nauce
Dostawca treści:
Biblioteka Nauki
Artykuł
Tytuł:
Research and applications of artificial neural networks in spatial analysis: Review
Autorzy:
Garczyńska, Ilona
Powiązania:
https://bibliotekanauki.pl/articles/29521035.pdf
Data publikacji:
2023
Wydawca:
Akademia Morska w Szczecinie. Wydawnictwo AMSz
Tematy:
spatial analysis
GIS
artificial neural network
artificial intelligence
geosciences
Opis:
The conducted review presents the possibility of using artificial neural networks in sectors related to environmental protection, agriculture, forestry, land uses, groundwater and bathymetric. Today there is a lot of research in these areas with different research methodologies. The result is the improvement of decision-making processes, design, and prediction of certain events that, with appropriate intervention, can prevent severe consequences for society. The review shows the capabilities to optimize and automate the processes of modeling urban and land dynamics. It examines the forecasts of assessment of the damage caused by natural phenomena. Detection of environmental changes via the analysis of certain time intervals and classification of objects on the basis of different images is presented. The practical aspects of this work include the ability to choose the correct artificial neural network model depending on the complexity of the problem. This factor is a novel element since previously reviewed articles did not encounter a study of the correlation between the chosen model or algorithm, depending on the use case or area of the problem. This article seeks to outline the reason for the interest in artificial intelligence. Its purpose is to find answers to the following questions: How can artificial neural networks be used for spatial analysis? What does the implementation of detailed algorithms depend on? It is proved that an artificial intelligence approach can be an effective and powerful tool in various domains where spatial aspects are important.
Źródło:
Zeszyty Naukowe Akademii Morskiej w Szczecinie; 2023, 74 (146); 35-45
1733-8670
2392-0378
Pojawia się w:
Zeszyty Naukowe Akademii Morskiej w Szczecinie
Dostawca treści:
Biblioteka Nauki
Artykuł
Tytuł:
Świat pod lupą fenomenologa
The world under the microscope of a phenomenologist
Autorzy:
Psica, Sylwia
Powiązania:
https://bibliotekanauki.pl/articles/690930.pdf
Data publikacji:
2017
Wydawca:
Copernicus Center Press
Tematy:
phenomenology
artificial intelligence
Opis:
Book review of: Aleksandra K. Przegalińska, Istoty wirtualne. Jak fenomenologia zmieniła sztuczną inteligencję, Wydawnictwo Universitas, Kraków 2016, ss. 268.
Recenzja książki: Aleksandra K. Przegalińska, Istoty wirtualne. Jak fenomenologia zmieniła sztuczną inteligencję, Wydawnictwo Universitas, Kraków 2016, ss. 268.
Źródło:
Zagadnienia Filozoficzne w Nauce; 2017, 63; 241-248
0867-8286
2451-0602
Pojawia się w:
Zagadnienia Filozoficzne w Nauce
Dostawca treści:
Biblioteka Nauki
Artykuł
Tytuł:
Potential and use of the googlenet ann for the purposes of inland water ships classification
Autorzy:
Bobkowska, Katarzyna
Bodus-Olkowska, Izabela
Powiązania:
https://bibliotekanauki.pl/articles/1573774.pdf
Data publikacji:
2020
Wydawca:
Politechnika Gdańska. Wydział Inżynierii Mechanicznej i Okrętownictwa
Tematy:
ship classification
image classification
geoinformatics
artificial intelligence
artificial neural network
Opis:
This article presents an analysis of the possibilities of using the pre-degraded GoogLeNet artificial neural network to classify inland vessels. Inland water authorities monitor the intensity of the vessels via CCTV. Such classification seems to be an improvement in their statutory tasks. The automatic classification of the inland vessels from video recording is a one of the main objectives of the Automatic Ship Recognition and Identification (SHREC) project. The image repository for the training purposes consists about 6,000 images of different categories of the vessels. Some images were gathered from internet websites, and some were collected by the project’s video cameras. The GoogLeNet network was trained and tested using 11 variants. These variants assumed modifications of image sets representing (e.g., change in the number of classes, change of class types, initial reconstruction of images, removal of images of insufficient quality). The final result of the classification quality was 83.6%. The newly obtained neural network can be an extension and a component of a comprehensive geoinformatics system for vessel recognition.
Źródło:
Polish Maritime Research; 2020, 4; 170-178
1233-2585
Pojawia się w:
Polish Maritime Research
Dostawca treści:
Biblioteka Nauki
Artykuł
Tytuł:
The survey of soft computing techniques for reliability prediction
Autorzy:
Smoczek, J.
Powiązania:
https://bibliotekanauki.pl/articles/246835.pdf
Data publikacji:
2012
Wydawca:
Instytut Techniczny Wojsk Lotniczych
Tematy:
reliability prediction
artificial intelligence
fuzzy logic
artificial neural network
genetic algorithm
Opis:
The objective of reliability prediction is to estimate a time of upcoming nonoperational state at the current operational state of a system through real-time monitoring operational parameters and/or performances. Hence, the predictive (proactive) maintenance in industrial systems involves operational conditions monitoring and online forecasting the useful life of machines equipment to support the decision-making process in selection of the best maintenance action to be carried out. The advanced warning of the failure possibility can bring the attention of machines operators and maintenance personnel to impending danger, and facilitate planning preventive and corrective operations, as well as inventory managing. This problem has been extensively studied in many scientific works, where the predictive models are based on the data-driven approaches that can be generally divided into statistical techniques (regression, ARMA models, Bayesian probability distribution estimation, etc.), grey system theory, and soft computing methods. The artificial intelligence is frequently addressed to the predictive problem by utilizing the learning capability of artificial neural network (ANN), and possibility of nonlinear mapping using fuzzy rules-based system (FRBS) or recognizing and optimizing data-derived pattern by using evolutionary algorithms. The paper is a survey of intelligent methods for failure prediction, and delivers the review of examples of scientific works presenting the computational intelligence-based approaches to predictive problem.
Źródło:
Journal of KONES; 2012, 19, 3; 407-414
1231-4005
2354-0133
Pojawia się w:
Journal of KONES
Dostawca treści:
Biblioteka Nauki
Artykuł
Tytuł:
Mobile robot navigation with the use of semantic map constructed from 3D laser range scans
Autorzy:
Siemiątkowska, B.
Szklarski, J.
Gnatowski, M.
Powiązania:
https://bibliotekanauki.pl/articles/206413.pdf
Data publikacji:
2011
Wydawca:
Polska Akademia Nauk. Instytut Badań Systemowych PAN
Tematy:
artificial intelligence
robotics
mapping
Opis:
We describe a system allowing a mobile robot equipped with a 3D laser range finder to navigate in the indoor and outdoor environment. A global map of the environment is constructed, and the particle filter algorithm is used in order to accurately determine the position of the robot. Based on data from the laser only, the robot is able to recognize certain classes of objects like a floor, a door, a washbasin, or a wastebasket, and places like corridors or rooms. For complex objects, the recognition process is based on the Haar feature identification. When an object is detected and identified, its position is associated with the appropriate place in the global map, making it possible to give orders to the robot with the use of semantic labels, e.g., "go to the nearest wastebasket ". The obstaclefree path is generated using a Cellular Neural Network, accounting for travel costs with distance or ground quality. This path planning method is fast and in comparison with the potential field method it does not suffer from the local minima problem. We present some results of experiments performed in a real indoor environment.
Źródło:
Control and Cybernetics; 2011, 40, 2; 437-453
0324-8569
Pojawia się w:
Control and Cybernetics
Dostawca treści:
Biblioteka Nauki
Artykuł
Tytuł:
Acciones para combatir el impacto del crimen en el ciberespacio. Prevención y detección con la Inteligencia Artificial
Actions to combat the impact of crime in cyberspace. Prevention and detection with Artificial Intelligence
Autorzy:
Guzmán, Silvia Irene Verdugo
Casteleiro, Ana Ochoa
Powiązania:
https://bibliotekanauki.pl/articles/2158851.pdf
Data publikacji:
2022
Wydawca:
Krakowska Akademia im. Andrzeja Frycza Modrzewskiego
Tematy:
artificial intelligence
darknet
cibercrime
hacker
cibersecurity
inteligencia artificial
internet oscura
cibercrimen
ciberdelincuente
ciberseguridad
Opis:
Los últimos dos siglos hemos visto nuevos tipos de delitos que han pasado a un primer plano, y que los delitos tradicionales han adoptado diferentes formas o un alcance completamente nuevo, especialmente digital. La especificidad de distintos ciberdelitos se encuentra en la internet profunda o Darknet, tales como el fraude informático, los ataques informáticos, abuso sexual y la explotación sexual de niños a través de sistemas informáticos etc. La aparición del nuevo coronavirus SARS-CoV-2 a fines del 2019 y la pandemia declarada por COVID-19 causada por el virus a comienzos del 2020, han servido para subrayar la importancia del combate al cibercrimen. La denuncia e investigación de delitos es crucial para las unidades policiales y las partes interesadas en la comprensión y análisis del ciberdelito. Entonces, gracias al desarrollo de la Inteligencia Artificial (I.A) y la recopilación de datos estadísticos son utilizados para realizar evaluaciones de delitos, castigos o sanciones, e identificar a las personas investigadas o perseguidas. La construcción de la cooperación en I.A con las partes interesadas en todo el sector público, la industria, la academia, así como las entidades de seguridad relacionadas, agencias de inteligencia etc., es un próximo paso esencial. También, cabe referirse a la ciberseguridad, que juega un papel importante en la defensa de las personas, de empresas e infraestructuras. Los proveedores de seguridad digital se enfocan en usuarios, dispositivos, redes, datos, aplicaciones o empresas para brindar soluciones con el fin de detectar y responder ante una amenaza cibernética.
Over the past two centuries, we have seen emerging types of crimes coming to the fore and traditional crimes taking on different forms or an entirely new scope altogether. Specificity of various cybercrimes are on the darknet, how computer fraud, computer attacks, the sexual abuse and sexual explotation of children through computer systems, etc. The emergence of the novel SARS-CoV-2 coronavirus in late 2019 and the ongoing COVID-19 pandemic caused by the virus on 2020, has served to underscore the importance of combat to cibercrime. The reporting and research of crimes is crucial for police units and stakeholders in understanding and analysis cybercrime. So, the I.A and collection of stadistical data information are used to evaluations of crime, punishments or sanctions and identifying the people investigated or persecuted. Building cooperation on I.A with stakeholders throughout the public sector, industry, academia, as well as related security entities, intelligence agencies and so on, is an essential next step. Also, the cybersecurity plays an important role in defending individuals, industry and infrastructure. Security vendors focus on users, network, data, application or entire enterprises to provide solutions to detect and respond to cyber threats.
Źródło:
Studia Prawnicze: rozprawy i materiały; 2022, (30), 1; 15-24
1689-8052
2451-0807
Pojawia się w:
Studia Prawnicze: rozprawy i materiały
Dostawca treści:
Biblioteka Nauki
Artykuł
Tytuł:
Las alternativas para una tributación de la robótica
The alternatives for a robotics taxation
Autorzy:
Parada, Luis Miguel Muleiro
Powiązania:
https://bibliotekanauki.pl/articles/2158865.pdf
Data publikacji:
2022
Wydawca:
Krakowska Akademia im. Andrzeja Frycza Modrzewskiego
Tematy:
inteligencia artificial
tributos
justicia
simplificación
artificial intelligence
robots
taxes
tax justice
tax simplification
Opis:
A estas alturas del siglo XXI parece deseable reflexionar desde el rigor científico sobre las posibilidades reales de incidir con técnicas impositivas que hagan contribuir específicamente a la robótica sin distorsionar ni complicar más todavía el sistema tributario. Una sociedad progresivamente robotizada en el entorno global digitalizado requiere plantearse la posibilidad de un gravamen en el marco de la inteligencia artificial aplicada. En los últimos años, diferentes voces han defendido la necesidad de alcanzar un tratamiento jurídico acorde a la nueva realidad en la que el imparable avance de la robótica está abriendo paso a una verdadera transformación social. En el ámbito tributario se han proyectado diversas alternativas, desde la posibilidad de regular un tributo referido al trabajo ejecutado por robots hasta exigir un gravamen específico o adicional por el uso y mantenimiento de los mismos. Pese a que las alternativas son bien diferentes la justicia tributaria debe mantenerse como prisma básico de evaluación.
In recent years different voices have defended the necessity to achieve a legal treatment according to the new reality, given that, the relentless advance of robotics is paving its path to a profound social transformation. A society, which is progressively automated in the digitised global environment, requires to formulate the possibility of a taxation within the applied artificial intelligence framework. Several alternatives have been proposed under the tax environment, from regulating a tax referred to the task executed by robots until demanding a specific or additional taxation on account of their use and maintenance. A priori, the alternatives are numerous but these required to be assessed from the perspective regarding the minimum determinants of the tax justice. In this paper, we will analyse what challenges the taxation deals, considering the necessary postulates that impose the fundamental guidelines of the tax order. The material principles of justice are reinterpreted in their content by their own digital revolution. These notions must clearly report the design of any suggestion and respect satisfactorily the configuration of the essential elements of a tax. At this stage in the 21st century, it seems desirable to reflect from the scientific rigour on the real possibilities of influencing by means of using tax techniques. These techniques could specifically contribute to robotics, but without distorting and complicating further the taxation system.
Źródło:
Studia Prawnicze: rozprawy i materiały; 2022, (30), 1; 105-115
1689-8052
2451-0807
Pojawia się w:
Studia Prawnicze: rozprawy i materiały
Dostawca treści:
Biblioteka Nauki
Artykuł
Tytuł:
Wykorzystanie techniki AdaBoost w modelach opartych na regresji z wykorzystaniem sztucznej inteligencji
Use of AdaBoost technique in regression based models using artificial intelligence
Autorzy:
Wołoszyn, Jacek
Powiązania:
https://bibliotekanauki.pl/articles/445996.pdf
Data publikacji:
2019
Wydawca:
Uniwersytet Rzeszowski
Tematy:
AdaBoost
sztuczna inteligencja
artificial intelligence
Opis:
Artykuł opisuje wykorzystanie sztucznej inteligencji w obliczeniach regresyjnych. Modele zbudowane w sposób tradycyjny oparte na klasycznych założeniach dostarczają możliwie precy-zyjnych informacji. Sam proces budowy modelu opiera się na wstępnym wyborze zmiennych wykorzystanych do jego tworzenia. Umiejętna selekcja zmiennych ma istotny wpływ na uzyskane parametry. Wykorzystując rozwiązanie przedstawione w tej pracy otrzymujemy model ze wstępnie dobranym optymalnym zbiorem treningowym. W kolejnych rozdziałach omówiono istotę analityki predykcyjnej, proces uczenia maszyno-wego, budowę drzewa decyzyjnego, pokazano przykład regresji wykorzystującego AdaBoost.
This article describes the use of artificial intelligence in regression calculations. Models built in a traditional manner based on classical assumptions provide the most precise information possi-ble. The model building process itself is based on the initial selection of variables used to create it. Skillful selection of variables has a significant impact on the obtained parameters. Using the solu-tion presented in this work, we get a model with a pre-selected optimal training set. The following chapters discuss the essence of predictive analytics, the machine learning pro-cess, the construction of a decision tree, an example of regression using AdaBoost.
Źródło:
Dydaktyka informatyki; 2019, 14; 162-169
2083-3156
Pojawia się w:
Dydaktyka informatyki
Dostawca treści:
Biblioteka Nauki
Artykuł
Tytuł:
Integration of product design and manufacturing with the use of artificial intelligent methods
Autorzy:
Kutschenreiter-Praszkiewicz, I.
Powiązania:
https://bibliotekanauki.pl/articles/99801.pdf
Data publikacji:
2011
Wydawca:
Wrocławska Rada Federacji Stowarzyszeń Naukowo-Technicznych
Tematy:
product design
manufacturing
artificial intelligence
Opis:
Product and process design is time- and cost consuming [15]. To reduce the cost of product development, it is necessary to integrate product and process design. The proposed product planning approach integrates activities involved in product design and manufacturing process. The aim of this paper is to develop a method of knowledge integration about customer needs, product and process characteristics. The range of analyses is limited to mechanical product type manufacturing for institutional customers. Customer needs are focused on functional characteristics of the product and the trade characteristics include product price, timing and warranty. Integration of functional requirements, product and process characteristics is needed to select the best product from a catalogue and adapt it to particular customer needs. From the given set of products, where a product is described by a set of attributes, the subset is chosen which roughly satisfies customer needs. Basing on artificial intelligent (AI) methods, data related to redesign and production processes is estimated.
Źródło:
Journal of Machine Engineering; 2011, 11, No. 1-2; 46-53
1895-7595
2391-8071
Pojawia się w:
Journal of Machine Engineering
Dostawca treści:
Biblioteka Nauki
Artykuł
Tytuł:
Evidence from Artificial Intelligence in General Administrative Procedure
Autorzy:
Dubiński, Łukasz
Powiązania:
https://bibliotekanauki.pl/articles/28672992.pdf
Data publikacji:
2022
Wydawca:
Uniwersytet im. Adama Mickiewicza w Poznaniu
Tematy:
artificial intelligence
evidence
administrative procedure
Opis:
Artificial intelligence is becoming an element of everyday life, and also a part of administrative proceedings. The legislator systematically adds regulations that allow public administration bodies to use artificial intelligence. At the same time, the Code of Administrative Procedure has not been amended in this respect. The article tries to establish what artificial intelligence evidence is. In addition, the issue of whether the rules on administrative evidence should be changed to cover the use of AI evidence is examined.
Źródło:
Adam Mickiewicz University Law Review; 2022, 14; 251-280
2450-0976
Pojawia się w:
Adam Mickiewicz University Law Review
Dostawca treści:
Biblioteka Nauki
Artykuł
Tytuł:
The identification of model parameters for a semi-empirical model of working process in the compression-ignition engine
Autorzy:
Brzozowski, K.
Nowakowski, J.
Powiązania:
https://bibliotekanauki.pl/articles/243901.pdf
Data publikacji:
2010
Wydawca:
Instytut Techniczny Wojsk Lotniczych
Tematy:
modelling
working cycle
artificial intelligence
Opis:
The paper concerns the identification process applied for a semi-empirical model of working process in the CI engine. The identification is based on pressure courses in the cylinder recorded during the experimental measurements on the test stand and is performed for estimation of values of the model parameters. Appropriate estimated values of the model parameters ensure minimization of the difference between measured and modelled pressure courses in the cylinder. The identification process can be divided into two stages. The first stage concerns identification of discrete values of the model parameters for a set of discrete engine operating conditions. The task of discrete identification is formulated as a dynamic optimisation task which is sohed using a genetic algorithm. The accuracy of the identification process is evaluated by comparison of measured and calculated values of main parameters which characterize the working cycle such as: the mean indicated pressure, thermal efficiency, the mean indicated pressure in working part of the cycle, maximal pressure of the cycle, the mass of the medium in the cylinder and the crankshaft angle for which the maximal pressure occurred. The second stage concerns generalization of the results for any technically possible engine operating conditions which is sohed by means of approximation. Feed-forward multilayer artificial neural networks are usedfor the approximation. The accuracy of the identification and some examples of verification of the model predictions are presented as well.
Źródło:
Journal of KONES; 2010, 17, 3; 41-48
1231-4005
2354-0133
Pojawia się w:
Journal of KONES
Dostawca treści:
Biblioteka Nauki
Artykuł
Tytuł:
Some effect of the unbalanced development of human intelligence: from trans-monkeys to trans-humans
Pewien skutek niezrównoważonego rozwoju ludzkiej inteligencji: od transmałp do transludzi
Autorzy:
Sztumski, Wiesław
Powiązania:
https://bibliotekanauki.pl/articles/2081649.pdf
Data publikacji:
2021-09-22
Wydawca:
Państwowa Wyższa Szkoła Zawodowa w Raciborzu
Tematy:
artificial intelligence
transhuman
transhumanity
posthuman
Opis:
The article is devoted to the effects of the unsustainable development of human intelligence that began with the emergence of artificial intelligence do by the imbalance between the slow pace of development of natural (biological) intelligence and the accelerate pace of development of artificial (machine) intelligence. First, in a breakthrough phase of the anthropogenesis process, the development of natural intelligence transformed hominids, which had exceeded their natural abilities, into transhominids, that is, into humans. Now, the development of artificial intelligence transforms humans, which exceed their natural (innate) abilities into so-called transhumans. The progress of artificial intelligence is making a real revolution not only in the field of technology, but also in the economy, medicine, agriculture, transport, finance, insurance, trade, education, work, etc. Thanks to this, our world and we in it are changing at a dizzying pace. Artificial intelligence has become strategically the most important technology of the twenty-first century. The country with the most advanced artificial Therefore, the superpowers accelerate the development of artificial intelligence in competing for global domination. This implies a huge increase in financial outlays, which then are missing for the implementation of other important tasks, such, as for example, to fight poverty. Along with the development of artificial intelligence, its alienation progresses and this causes the dehumanization of people. The author presents a number of benefits and threats resulting from the use of artificial intelligence, which do not balance. Among the greatest threats, they are transhuman and artificial super-intelligence embodied in a super- robot. “The full development of artificial intelligence could spell the end of the human race. Man, who is limited by the slow evolution of biological intelligence, will not be able to compete with the transhuman, whose artificial intelligence is developing faster. He needs to be replaced by a transhuman”.1 Artificial intelligence is able to make man equal to God. It is not difficult to imagine what would happen if such a robot became the "god" who would rule the world and the cosmos, and on whose decisions or whims the fate of individuals, humanity, the world and the survival of our species would depend.
Źródło:
Eunomia – Rozwój Zrównoważony – Sustainable Development; 2021, 1(100); 159-170
1897-2349
2657-5760
Pojawia się w:
Eunomia – Rozwój Zrównoważony – Sustainable Development
Dostawca treści:
Biblioteka Nauki
Artykuł
Tytuł:
Comparison of an effectiveness of artificial neural networks for various activation functions
Porównanie efektywności sztucznych sieci neuronowych dla różnych funkcji aktywacji
Autorzy:
Florek, Daniel
Miłosz, Marek
Powiązania:
https://bibliotekanauki.pl/articles/24083568.pdf
Data publikacji:
2023
Wydawca:
Politechnika Lubelska. Instytut Informatyki
Tematy:
activation functions
artificial neural networks
artificial intelligence
funkcja aktywacji
sztuczne sieci neuronowe
sztuczna inteligencja
Opis:
Activation functions play an important role in artificial neural networks (ANNs) because they break the linearity in the data transformations that are performed by models. Thanks to the recent spike in interest around the topic of ANNs, new improvements to activation functions are emerging. The paper presents the results of research on the effectiveness of ANNs for ReLU, Leaky ReLU, ELU, and Swish activation functions. Four different data sets, and three different network architectures were used. Results show that Leaky ReLU, ELU and Swish functions work better in deep and more complex architectures which are to alleviate vanishing gradient and dead neurons problems. Neither of the three aforementioned functions comes ahead in accuracy in all used datasets, although Swish activation speeds up training considerably and ReLU is the fastest during prediction process.
Funkcje aktywacji, przełamując linową naturę transformacji zachodzących w sztucznych sieciach neuronowych (SSN), pozwalają na uczenie skomplikowanych wzorców występujących w danych wejściowych, np. w obrazach. Wzrost zain-teresowania wokół SSN skłonił naukowców do badań wokół różnolitych aktywacji, które mogą dać przewagę podczas uczenia jak i przewidywania, ostatecznie przyczyniając się do powstania nowych, interesujących rozwiązań. W artykule przedstawiono wyniki badań nad efektywnością SSN dla funkcji ReLU, Leaky ReLU, ELU oraz Swish, przy użyciu czterech zbiorów danych i trzech różnych architektur SSN. Wyniki pokazują, że funkcje Leaky ReLU, ELU i Swish lepiej sprawdzają się w głębokich i bardziej skomplikowanych architekturach, mając za zadanie zapobieganie proble-mom zanikającego gradientu (ang. Vanishing Gradient) i martwych neuronów (ang. Dead neurons). Żadna z trzech wyżej wymienionych funkcji nie ma przewagi w celności (ang. Accuracy), jednakże Swish znacznie przyspiesza ucze-nie SSN, a ReLU jest najszybsza w procesie przewidywania
Źródło:
Journal of Computer Sciences Institute; 2023, 26; 7--12
2544-0764
Pojawia się w:
Journal of Computer Sciences Institute
Dostawca treści:
Biblioteka Nauki
Artykuł
Wyświetlanie 1-15 z 562

Ta witryna wykorzystuje pliki cookies do przechowywania informacji na Twoim komputerze. Pliki cookies stosujemy w celu świadczenia usług na najwyższym poziomie, w tym w sposób dostosowany do indywidualnych potrzeb. Korzystanie z witryny bez zmiany ustawień dotyczących cookies oznacza, że będą one zamieszczane w Twoim komputerze. W każdym momencie możesz dokonać zmiany ustawień dotyczących cookies