Number of the records: 1
Mappings between High-Dimensional Representations in Connectionistic Systems
- 1.0103262 - UIVT-O 20040003 RIV SK eng C - Conference Paper (international conference)
Kůrková, Věra
Mappings between High-Dimensional Representations in Connectionistic Systems.
[Transformace vysoko dimenzionálních reprezentací v konekcionistických systémech.]
Machine Intelligence Quo Vadis? New Jersey: World Scientific, 2004 - (Sinčák, P.; Vaščák, J.; Hirota, K.), s. 31-45. Advances in Fuzzy Systems - Applications and Theory, 21. ISBN 981-238-751-X.
[European Symposium on Computational Intelligence. Košice (SK), 16.06.2002-19.06.2002]
R&D Projects: GA ČR GA201/00/1489
Institutional research plan: CEZ:AV0Z1030915
Keywords : approximations of high-dimensional functions * perceptron networks * variation norm
Subject RIV: BA - General Mathematics
DOI: https://doi.org/10.1142/9789812562531_0002
Approximations of high-dimensional mappings by neural network is investigated in the context of nonlinear approximation theory. It is shown that the "curse of dimensionality" can be avoided when certain norms of mappings are kept low. There are described properties and methods of derivation of estimates of such norms. The results are applied to perceptron and RBF networks.
Aproximace vysoko dimenzionálních zobrazení pomocí neuronových sítí je studována v kontextu nelineární aproximační teorie. Je ukázáno, že je možné se vyhnout tzv. prokletí dimenzionality pokud jsou určité normy zobrazení nízké. Jsou popsány vlastnosti a metody odhadu těchto norem. Výsledky jsou aplikovány na RBF a perceptronové sítě.
Permanent Link: http://hdl.handle.net/11104/0010574
Number of the records: 1