Self-Distillation for Randomized Neural Networks

<p dir="ltr">Knowledge distillation (KD) is a conventional method in the field of deep learning that enables the transfer of dark knowledge from a teacher model to a student model, consequently improving the performance of the student model. In randomized neural networks, due to the...

وصف كامل

محفوظ في:
التفاصيل البيبلوغرافية
المؤلف الرئيسي: Minghui Hu (2457952) (author)
مؤلفون آخرون: Ruobin Gao (16003195) (author), Ponnuthurai Nagaratnam Suganthan (11274636) (author)
منشور في: 2023
الموضوعات:
الوسوم: إضافة وسم
لا توجد وسوم, كن أول من يضع وسما على هذه التسجيلة!
_version_ 1864513527006363648
author Minghui Hu (2457952)
author2 Ruobin Gao (16003195)
Ponnuthurai Nagaratnam Suganthan (11274636)
author2_role author
author
author_facet Minghui Hu (2457952)
Ruobin Gao (16003195)
Ponnuthurai Nagaratnam Suganthan (11274636)
author_role author
dc.creator.none.fl_str_mv Minghui Hu (2457952)
Ruobin Gao (16003195)
Ponnuthurai Nagaratnam Suganthan (11274636)
dc.date.none.fl_str_mv 2023-08-16T09:00:00Z
dc.identifier.none.fl_str_mv 10.1109/tnnls.2023.3292063
dc.relation.none.fl_str_mv https://figshare.com/articles/journal_contribution/Self-Distillation_for_Randomized_Neural_Networks/25243387
dc.rights.none.fl_str_mv CC BY 4.0
info:eu-repo/semantics/openAccess
dc.subject.none.fl_str_mv Information and computing sciences
Artificial intelligence
Biological neural networks
Knowledge engineering
Neurons
Training
Pipelines
Closed-form solutions
Predictive models
Knowledge distillation (KD)
random vector functional link (RVFL)
randomized neural network
self-distillation
dc.title.none.fl_str_mv Self-Distillation for Randomized Neural Networks
dc.type.none.fl_str_mv Text
Journal contribution
info:eu-repo/semantics/publishedVersion
text
contribution to journal
description <p dir="ltr">Knowledge distillation (KD) is a conventional method in the field of deep learning that enables the transfer of dark knowledge from a teacher model to a student model, consequently improving the performance of the student model. In randomized neural networks, due to the simple topology of network architecture and the insignificant relationship between model performance and model size, KD is not able to improve model performance. In this work, we propose a self-distillation pipeline for randomized neural networks: the predictions of the network itself are regarded as the additional target, which are mixed with the weighted original target as a distillation target containing dark knowledge to supervise the training of the model. All the predictions during multi-generation self-distillation process can be integrated by a multi-teacher method. By induction, we have additionally arrived at the methods for infinite self-distillation (ISD) of randomized neural networks. We then provide relevant theoretical analysis about the self-distillation method for randomized neural networks. Furthermore, we demonstrated the effectiveness of the proposed method in practical applications on several benchmark datasets.</p><h2>Other Information</h2><p dir="ltr">Published in: IEEE Transactions on Neural Networks and Learning Systems<br>License: <a href="https://creativecommons.org/licenses/by/4.0/" target="_blank">https://creativecommons.org/licenses/by/4.0/</a><br>See article on publisher's website: <a href="https://dx.doi.org/10.1109/tnnls.2023.3292063" target="_blank">https://dx.doi.org/10.1109/tnnls.2023.3292063</a></p>
eu_rights_str_mv openAccess
id Manara2_e3bbfbb0ac09bb3bd7fe43247b43c1de
identifier_str_mv 10.1109/tnnls.2023.3292063
network_acronym_str Manara2
network_name_str Manara2
oai_identifier_str oai:figshare.com:article/25243387
publishDate 2023
repository.mail.fl_str_mv
repository.name.fl_str_mv
repository_id_str
rights_invalid_str_mv CC BY 4.0
spelling Self-Distillation for Randomized Neural NetworksMinghui Hu (2457952)Ruobin Gao (16003195)Ponnuthurai Nagaratnam Suganthan (11274636)Information and computing sciencesArtificial intelligenceBiological neural networksKnowledge engineeringNeuronsTrainingPipelinesClosed-form solutionsPredictive modelsKnowledge distillation (KD)random vector functional link (RVFL)randomized neural networkself-distillation<p dir="ltr">Knowledge distillation (KD) is a conventional method in the field of deep learning that enables the transfer of dark knowledge from a teacher model to a student model, consequently improving the performance of the student model. In randomized neural networks, due to the simple topology of network architecture and the insignificant relationship between model performance and model size, KD is not able to improve model performance. In this work, we propose a self-distillation pipeline for randomized neural networks: the predictions of the network itself are regarded as the additional target, which are mixed with the weighted original target as a distillation target containing dark knowledge to supervise the training of the model. All the predictions during multi-generation self-distillation process can be integrated by a multi-teacher method. By induction, we have additionally arrived at the methods for infinite self-distillation (ISD) of randomized neural networks. We then provide relevant theoretical analysis about the self-distillation method for randomized neural networks. Furthermore, we demonstrated the effectiveness of the proposed method in practical applications on several benchmark datasets.</p><h2>Other Information</h2><p dir="ltr">Published in: IEEE Transactions on Neural Networks and Learning Systems<br>License: <a href="https://creativecommons.org/licenses/by/4.0/" target="_blank">https://creativecommons.org/licenses/by/4.0/</a><br>See article on publisher's website: <a href="https://dx.doi.org/10.1109/tnnls.2023.3292063" target="_blank">https://dx.doi.org/10.1109/tnnls.2023.3292063</a></p>2023-08-16T09:00:00ZTextJournal contributioninfo:eu-repo/semantics/publishedVersiontextcontribution to journal10.1109/tnnls.2023.3292063https://figshare.com/articles/journal_contribution/Self-Distillation_for_Randomized_Neural_Networks/25243387CC BY 4.0info:eu-repo/semantics/openAccessoai:figshare.com:article/252433872023-08-16T09:00:00Z
spellingShingle Self-Distillation for Randomized Neural Networks
Minghui Hu (2457952)
Information and computing sciences
Artificial intelligence
Biological neural networks
Knowledge engineering
Neurons
Training
Pipelines
Closed-form solutions
Predictive models
Knowledge distillation (KD)
random vector functional link (RVFL)
randomized neural network
self-distillation
status_str publishedVersion
title Self-Distillation for Randomized Neural Networks
title_full Self-Distillation for Randomized Neural Networks
title_fullStr Self-Distillation for Randomized Neural Networks
title_full_unstemmed Self-Distillation for Randomized Neural Networks
title_short Self-Distillation for Randomized Neural Networks
title_sort Self-Distillation for Randomized Neural Networks
topic Information and computing sciences
Artificial intelligence
Biological neural networks
Knowledge engineering
Neurons
Training
Pipelines
Closed-form solutions
Predictive models
Knowledge distillation (KD)
random vector functional link (RVFL)
randomized neural network
self-distillation