Evolution Of Activation Functions for Neural Architecture Search

The introduction of the ReLU function in neural network architectures yielded substantial improvements over sigmoidal activation functions and allowed for the training of deep networks. Ever since, the search for new activation functions in neural networks has been an active research topic. However,...

وصف كامل

محفوظ في:
التفاصيل البيبلوغرافية
المؤلف الرئيسي: Nader, Andrew (author)
التنسيق: masterThesis
منشور في: 2020
الموضوعات:
الوصول للمادة أونلاين:http://hdl.handle.net/10725/13847
https://doi.org/10.26756/th.2022.373
http://libraries.lau.edu.lb/research/laur/terms-of-use/thesis.php
الوسوم: إضافة وسم
لا توجد وسوم, كن أول من يضع وسما على هذه التسجيلة!
_version_ 1864513468321759232
author Nader, Andrew
author_facet Nader, Andrew
author_role author
dc.creator.none.fl_str_mv Nader, Andrew
dc.date.none.fl_str_mv 2020
2020-05-18
2022-07-21T08:40:45Z
2022-07-21T08:40:45Z
dc.identifier.none.fl_str_mv http://hdl.handle.net/10725/13847
https://doi.org/10.26756/th.2022.373
http://libraries.lau.edu.lb/research/laur/terms-of-use/thesis.php
dc.language.none.fl_str_mv en
dc.publisher.none.fl_str_mv Lebanese American University
dc.rights.*.fl_str_mv info:eu-repo/semantics/openAccess
dc.subject.none.fl_str_mv Computer network architectures
Neural networks (Computer science)
Machine learning
Lebanese American University -- Dissertations
Dissertations, Academic
dc.title.none.fl_str_mv Evolution Of Activation Functions for Neural Architecture Search
dc.type.none.fl_str_mv Thesis
info:eu-repo/semantics/publishedVersion
info:eu-repo/semantics/masterThesis
description The introduction of the ReLU function in neural network architectures yielded substantial improvements over sigmoidal activation functions and allowed for the training of deep networks. Ever since, the search for new activation functions in neural networks has been an active research topic. However, to the best of our knowledge, the design of new activation functions has mostly been done by hand. In this work, we propose the use of a self-adaptive evolutionary algorithm that searches for new activation functions using a genetic programming approach, and we compare the performance of the obtained activation functions to ReLU. We also analyze the shape of the obtained activations to see if they have any common traits such as monotonicity or piece-wise linearity, and we study the effects of the self-adaptation to see which operators perform well in the context of a search for new activation functions. We perform a thorough experimental study on datasets of different sizes and types, using different types of neural network architectures. We report favorable results obtained from the mean and standard deviation of the performance metrics over multiple runs.
eu_rights_str_mv openAccess
format masterThesis
id LAURepo_664f3edea6c5fa156fae77d249aba407
language_invalid_str_mv en
network_acronym_str LAURepo
network_name_str Lebanese American University repository
oai_identifier_str oai:laur.lau.edu.lb:10725/13847
publishDate 2020
publisher.none.fl_str_mv Lebanese American University
repository.mail.fl_str_mv
repository.name.fl_str_mv
repository_id_str
spelling Evolution Of Activation Functions for Neural Architecture SearchNader, AndrewComputer network architecturesNeural networks (Computer science)Machine learningLebanese American University -- DissertationsDissertations, AcademicThe introduction of the ReLU function in neural network architectures yielded substantial improvements over sigmoidal activation functions and allowed for the training of deep networks. Ever since, the search for new activation functions in neural networks has been an active research topic. However, to the best of our knowledge, the design of new activation functions has mostly been done by hand. In this work, we propose the use of a self-adaptive evolutionary algorithm that searches for new activation functions using a genetic programming approach, and we compare the performance of the obtained activation functions to ReLU. We also analyze the shape of the obtained activations to see if they have any common traits such as monotonicity or piece-wise linearity, and we study the effects of the self-adaptation to see which operators perform well in the context of a search for new activation functions. We perform a thorough experimental study on datasets of different sizes and types, using different types of neural network architectures. We report favorable results obtained from the mean and standard deviation of the performance metrics over multiple runs.1 online resource (xii, 108 leaves): col. ill.Bibliography: leaf 98-108.Lebanese American University2022-07-21T08:40:45Z2022-07-21T08:40:45Z20202020-05-18Thesisinfo:eu-repo/semantics/publishedVersioninfo:eu-repo/semantics/masterThesishttp://hdl.handle.net/10725/13847https://doi.org/10.26756/th.2022.373http://libraries.lau.edu.lb/research/laur/terms-of-use/thesis.phpeninfo:eu-repo/semantics/openAccessoai:laur.lau.edu.lb:10725/138472022-07-21T08:41:37Z
spellingShingle Evolution Of Activation Functions for Neural Architecture Search
Nader, Andrew
Computer network architectures
Neural networks (Computer science)
Machine learning
Lebanese American University -- Dissertations
Dissertations, Academic
status_str publishedVersion
title Evolution Of Activation Functions for Neural Architecture Search
title_full Evolution Of Activation Functions for Neural Architecture Search
title_fullStr Evolution Of Activation Functions for Neural Architecture Search
title_full_unstemmed Evolution Of Activation Functions for Neural Architecture Search
title_short Evolution Of Activation Functions for Neural Architecture Search
title_sort Evolution Of Activation Functions for Neural Architecture Search
topic Computer network architectures
Neural networks (Computer science)
Machine learning
Lebanese American University -- Dissertations
Dissertations, Academic
url http://hdl.handle.net/10725/13847
https://doi.org/10.26756/th.2022.373
http://libraries.lau.edu.lb/research/laur/terms-of-use/thesis.php