Please use this identifier to cite or link to this item: http://ir.juit.ac.in:8080/jspui/jspui/handle/123456789/5307
Full metadata record
DC FieldValueLanguage
dc.contributor.authorPatel, Lokendra Singh-
dc.contributor.authorSaha, Suman [Guided by]-
dc.date.accessioned2022-07-28T15:47:32Z-
dc.date.available2022-07-28T15:47:32Z-
dc.date.issued2015-
dc.identifier.urihttp://ir.juit.ac.in:8080/jspui//xmlui/handle/123456789/5307-
dc.description.abstractLarge data sets have tens of thousands to millions of training instances, which suffers from high time and space complexity. To reduce the time and space complexity, we propose efficient Nystrom method to approximate kernel matrix, which is used in many machine learning methods such as kernel-based methods, e.g. Kernel Ridge Regression, Kernel Principle Component Analysis and Support Vector Machine. This thesis focuses on sampling based matrix approximation methods. Matrix approximation will help to speed up the kernel based algorithms to large data set. We give the desirable error bound both in the Frobenius and spectral norm for the quality of approximation. Based on these error bounds, we analyze the quality of approximation in kernel based algorithms. We present guarantees on approximation accuracy based on various matrix properties and analyze the effect of matrix approximation on actual kernel-based algorithms. Our proposed algorithm gives the lower error bound for the low rank approximation of the kernel matrixen_US
dc.language.isoenen_US
dc.publisherJaypee University of Information Technology, Solan, H.P.en_US
dc.subjectLinear algebraen_US
dc.subjectVector terminologyen_US
dc.subjectMatrix approximationen_US
dc.subjectLinear kernelen_US
dc.subjectTruncated SVDen_US
dc.subjectAlgorithmen_US
dc.titleMatrix Approximation Algorithms and Its Applicationsen_US
dc.typeProject Reporten_US
Appears in Collections:Dissertations (M.Tech.)

Files in This Item:
File Description SizeFormat 
Matrix Approximation Algorithms and Its Applications.pdf1.12 MBAdobe PDFView/Open


Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.