Neural Networks as a Convex Problem

A Master of Science thesis in Mathematics by Baha Khalil entitled, "Neural Networks as a Convex Problem," submitted in September 2016. Thesis advisor is Dr. Dmitry Efimov. Soft and hard copy available.

محفوظ في:
التفاصيل البيبلوغرافية
المؤلف الرئيسي: Khalil, Baha (author)
التنسيق: doctoralThesis
منشور في: 2016
الموضوعات:
الوصول للمادة أونلاين:http://hdl.handle.net/11073/8493
الوسوم: إضافة وسم
لا توجد وسوم, كن أول من يضع وسما على هذه التسجيلة!
_version_ 1864513442004598784
author Khalil, Baha
author_facet Khalil, Baha
author_role author
dc.contributor.none.fl_str_mv Efimov, Dmitry
dc.creator.none.fl_str_mv Khalil, Baha
dc.date.none.fl_str_mv 2016-10-10T05:57:32Z
2016-10-10T05:57:32Z
2016-09
dc.format.none.fl_str_mv application/pdf
dc.identifier.none.fl_str_mv 29.232-2016.08
http://hdl.handle.net/11073/8493
dc.language.none.fl_str_mv en_US
dc.subject.none.fl_str_mv Neural Networks
Convex Optimization
Gauss-Newton Matrix
Neural networks (Computer science)
Mathematical optimization
Convex functions
dc.title.none.fl_str_mv Neural Networks as a Convex Problem
dc.type.none.fl_str_mv info:eu-repo/semantics/publishedVersion
info:eu-repo/semantics/doctoralThesis
description A Master of Science thesis in Mathematics by Baha Khalil entitled, "Neural Networks as a Convex Problem," submitted in September 2016. Thesis advisor is Dr. Dmitry Efimov. Soft and hard copy available.
format doctoralThesis
id aus_5226dce90ff0f32b3cda331b6a3f0a5c
identifier_str_mv 29.232-2016.08
language_invalid_str_mv en_US
network_acronym_str aus
network_name_str aus
oai_identifier_str oai:repository.aus.edu:11073/8493
publishDate 2016
repository.mail.fl_str_mv
repository.name.fl_str_mv
repository_id_str
spelling Neural Networks as a Convex ProblemKhalil, BahaNeural NetworksConvex OptimizationGauss-Newton MatrixNeural networks (Computer science)Mathematical optimizationConvex functionsA Master of Science thesis in Mathematics by Baha Khalil entitled, "Neural Networks as a Convex Problem," submitted in September 2016. Thesis advisor is Dr. Dmitry Efimov. Soft and hard copy available.We reformulated the problem of training the neural networks model into a convex optimization problem by performing a local quadratic expansion of the cost function and adding the necessary constraints. We designed a new algorithm that extends the back propagation algorithm for parameters estimation by using second-order optimization methods. We computed the second order mixed partial derivatives of the cost function for a single hidden layer neural network model to construct the Hessian matrix. We used the Gauss-Newton approximation instead of the Hessian matrix to avoid the analytical computation of the second order derivative terms for higher order neural network topologies. To compare the accuracy and computational complexity of our proposed algorithm versus the standard back propagation we tested both algorithms in different applications, such as: regression, classification, and ranking.College of Arts and SciencesDepartment of Mathematics and StatisticsMaster of Science in Mathematics (MSMTH)Efimov, Dmitry2016-10-10T05:57:32Z2016-10-10T05:57:32Z2016-09info:eu-repo/semantics/publishedVersioninfo:eu-repo/semantics/doctoralThesisapplication/pdf29.232-2016.08http://hdl.handle.net/11073/8493en_USoai:repository.aus.edu:11073/84932025-06-26T12:25:59Z
spellingShingle Neural Networks as a Convex Problem
Khalil, Baha
Neural Networks
Convex Optimization
Gauss-Newton Matrix
Neural networks (Computer science)
Mathematical optimization
Convex functions
status_str publishedVersion
title Neural Networks as a Convex Problem
title_full Neural Networks as a Convex Problem
title_fullStr Neural Networks as a Convex Problem
title_full_unstemmed Neural Networks as a Convex Problem
title_short Neural Networks as a Convex Problem
title_sort Neural Networks as a Convex Problem
topic Neural Networks
Convex Optimization
Gauss-Newton Matrix
Neural networks (Computer science)
Mathematical optimization
Convex functions
url http://hdl.handle.net/11073/8493