o:1790 Algoritmi za brzo aproksimativno spektralno učenje Algorithms for fast approximate spectral learning: doctoral dissertation : doctoral dissertation sr This thesis presents learning algorithms which use theinformation stored in the spectrum (eigenvalues andeigenvectors) of a matrix derived from the input set. Matricesin question are graph matrices or kernel matrices. However, thealgorithms which use these matrices have either a quadratic orcubic time complexity and quadratic memory complexity.Therefore, in this thesis the algorithms will be presented thatapproximate those matrices and reduce the time and memorycomplexity to the linear one. Also, these algorithms will becompared with the other algorithms that solve this problem, andtheir empirical and theoretical analysis will be presented. Biobibliografija: list. 114-115;Bibliografija: list. 108-113. Datum odbrane: Artificial Intelligence; Machine Learning 91552100 56370441 91552101 8522 2022-11-18T10:08:26.053Z 45 no 46 Aleksandar B., 1989- Trokicić 79656969 2021 63 mentor Branimir, 1967- Todorović 28413287 2021 63 član komisije Miroslav Ćirić 2021 63 član komisije Zoran Ognjanović 2021 63 član komisije Dragan Janković 2021 63 član komisije Marko Petković 2021 115, [10] listova 3005214 http://phaidrani.ni.ac.rs/o:1790 no yes 12 2022-11-18T10:08:26.320Z 70 klasterovanje, kernel regresija, spektralne metode,aproksimacija, Nistromova metoda, Laplasova matrica clustering, kernel regression, spectral methods, approximation,Nystrom method, Laplacian matrix 004.8(043.3) 1738 18A07 18A0701 2021