Customizing Synthetic Data for Data-Free Student Learning

Data-free knowledge distillation (DFKD) aims to obtain a lightweight student model without original training data. Existing works generally synthesize data from the pre-trained teacher model to replace the original training data for student learning. To more effectively train the student model, the...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Luo, Shiya, Chen, Defang, Wang, Can
Format: Artikel
Sprache:eng
Schlagworte:
Online-Zugang:Volltext bestellen
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
container_end_page
container_issue
container_start_page
container_title
container_volume
creator Luo, Shiya
Chen, Defang
Wang, Can
description Data-free knowledge distillation (DFKD) aims to obtain a lightweight student model without original training data. Existing works generally synthesize data from the pre-trained teacher model to replace the original training data for student learning. To more effectively train the student model, the synthetic data shall be customized to the current student learning ability. However, this is ignored in the existing DFKD methods and thus negatively affects the student training. To address this issue, we propose Customizing Synthetic Data for Data-Free Student Learning (CSD) in this paper, which achieves adaptive data synthesis using a self-supervised augmented auxiliary task to estimate the student learning ability. Specifically, data synthesis is dynamically adjusted to enlarge the cross entropy between the labels and the predictions from the self-supervised augmented task, thus generating hard samples for the student model. The experiments on various datasets and teacher-student models show the effectiveness of our proposed method. Code is available at: $\href{https://github.com/luoshiya/CSD}{https://github.com/luoshiya/CSD}$
doi_str_mv 10.48550/arxiv.2307.04542
format Article
fullrecord <record><control><sourceid>arxiv_GOX</sourceid><recordid>TN_cdi_arxiv_primary_2307_04542</recordid><sourceformat>XML</sourceformat><sourcesystem>PC</sourcesystem><sourcerecordid>2307_04542</sourcerecordid><originalsourceid>FETCH-LOGICAL-a672-b8cb2578014b0dd6cc63c18dd012dac131c4d58dde2bb27ba696e06d71b146293</originalsourceid><addsrcrecordid>eNotj8uqwjAYhLNxIR4fwJV5gdbc0y4P9QoFF7ovfy56AscqMYr69Gp1NTMwDPMhNKIkF4WUZALxFq4540TnREjB-qisLud0PIRHaPd4c2_Tn0_B4ikkwLtj7Ew2j97jTbo43yZce4jtq_2Dejv4P_vhVwdoO59tq2VWrxer6rfOQGmWmcIaJnVBqDDEOWWt4pYWzhHKHFjKqRVOvrJnxjBtQJXKE-U0NVQoVvIBGn9mu-_NKYYDxHvzZmg6Bv4E0NBBEg</addsrcrecordid><sourcetype>Open Access Repository</sourcetype><iscdi>true</iscdi><recordtype>article</recordtype></control><display><type>article</type><title>Customizing Synthetic Data for Data-Free Student Learning</title><source>arXiv.org</source><creator>Luo, Shiya ; Chen, Defang ; Wang, Can</creator><creatorcontrib>Luo, Shiya ; Chen, Defang ; Wang, Can</creatorcontrib><description>Data-free knowledge distillation (DFKD) aims to obtain a lightweight student model without original training data. Existing works generally synthesize data from the pre-trained teacher model to replace the original training data for student learning. To more effectively train the student model, the synthetic data shall be customized to the current student learning ability. However, this is ignored in the existing DFKD methods and thus negatively affects the student training. To address this issue, we propose Customizing Synthetic Data for Data-Free Student Learning (CSD) in this paper, which achieves adaptive data synthesis using a self-supervised augmented auxiliary task to estimate the student learning ability. Specifically, data synthesis is dynamically adjusted to enlarge the cross entropy between the labels and the predictions from the self-supervised augmented task, thus generating hard samples for the student model. The experiments on various datasets and teacher-student models show the effectiveness of our proposed method. Code is available at: $\href{https://github.com/luoshiya/CSD}{https://github.com/luoshiya/CSD}$</description><identifier>DOI: 10.48550/arxiv.2307.04542</identifier><language>eng</language><subject>Computer Science - Computer Vision and Pattern Recognition</subject><creationdate>2023-07</creationdate><rights>http://arxiv.org/licenses/nonexclusive-distrib/1.0</rights><oa>free_for_read</oa><woscitedreferencessubscribed>false</woscitedreferencessubscribed></display><links><openurl>$$Topenurl_article</openurl><openurlfulltext>$$Topenurlfull_article</openurlfulltext><thumbnail>$$Tsyndetics_thumb_exl</thumbnail><link.rule.ids>228,230,776,881</link.rule.ids><linktorsrc>$$Uhttps://arxiv.org/abs/2307.04542$$EView_record_in_Cornell_University$$FView_record_in_$$GCornell_University$$Hfree_for_read</linktorsrc><backlink>$$Uhttps://doi.org/10.48550/arXiv.2307.04542$$DView paper in arXiv$$Hfree_for_read</backlink></links><search><creatorcontrib>Luo, Shiya</creatorcontrib><creatorcontrib>Chen, Defang</creatorcontrib><creatorcontrib>Wang, Can</creatorcontrib><title>Customizing Synthetic Data for Data-Free Student Learning</title><description>Data-free knowledge distillation (DFKD) aims to obtain a lightweight student model without original training data. Existing works generally synthesize data from the pre-trained teacher model to replace the original training data for student learning. To more effectively train the student model, the synthetic data shall be customized to the current student learning ability. However, this is ignored in the existing DFKD methods and thus negatively affects the student training. To address this issue, we propose Customizing Synthetic Data for Data-Free Student Learning (CSD) in this paper, which achieves adaptive data synthesis using a self-supervised augmented auxiliary task to estimate the student learning ability. Specifically, data synthesis is dynamically adjusted to enlarge the cross entropy between the labels and the predictions from the self-supervised augmented task, thus generating hard samples for the student model. The experiments on various datasets and teacher-student models show the effectiveness of our proposed method. Code is available at: $\href{https://github.com/luoshiya/CSD}{https://github.com/luoshiya/CSD}$</description><subject>Computer Science - Computer Vision and Pattern Recognition</subject><fulltext>true</fulltext><rsrctype>article</rsrctype><creationdate>2023</creationdate><recordtype>article</recordtype><sourceid>GOX</sourceid><recordid>eNotj8uqwjAYhLNxIR4fwJV5gdbc0y4P9QoFF7ovfy56AscqMYr69Gp1NTMwDPMhNKIkF4WUZALxFq4540TnREjB-qisLud0PIRHaPd4c2_Tn0_B4ikkwLtj7Ew2j97jTbo43yZce4jtq_2Dejv4P_vhVwdoO59tq2VWrxer6rfOQGmWmcIaJnVBqDDEOWWt4pYWzhHKHFjKqRVOvrJnxjBtQJXKE-U0NVQoVvIBGn9mu-_NKYYDxHvzZmg6Bv4E0NBBEg</recordid><startdate>20230710</startdate><enddate>20230710</enddate><creator>Luo, Shiya</creator><creator>Chen, Defang</creator><creator>Wang, Can</creator><scope>AKY</scope><scope>GOX</scope></search><sort><creationdate>20230710</creationdate><title>Customizing Synthetic Data for Data-Free Student Learning</title><author>Luo, Shiya ; Chen, Defang ; Wang, Can</author></sort><facets><frbrtype>5</frbrtype><frbrgroupid>cdi_FETCH-LOGICAL-a672-b8cb2578014b0dd6cc63c18dd012dac131c4d58dde2bb27ba696e06d71b146293</frbrgroupid><rsrctype>articles</rsrctype><prefilter>articles</prefilter><language>eng</language><creationdate>2023</creationdate><topic>Computer Science - Computer Vision and Pattern Recognition</topic><toplevel>online_resources</toplevel><creatorcontrib>Luo, Shiya</creatorcontrib><creatorcontrib>Chen, Defang</creatorcontrib><creatorcontrib>Wang, Can</creatorcontrib><collection>arXiv Computer Science</collection><collection>arXiv.org</collection></facets><delivery><delcategory>Remote Search Resource</delcategory><fulltext>fulltext_linktorsrc</fulltext></delivery><addata><au>Luo, Shiya</au><au>Chen, Defang</au><au>Wang, Can</au><format>journal</format><genre>article</genre><ristype>JOUR</ristype><atitle>Customizing Synthetic Data for Data-Free Student Learning</atitle><date>2023-07-10</date><risdate>2023</risdate><abstract>Data-free knowledge distillation (DFKD) aims to obtain a lightweight student model without original training data. Existing works generally synthesize data from the pre-trained teacher model to replace the original training data for student learning. To more effectively train the student model, the synthetic data shall be customized to the current student learning ability. However, this is ignored in the existing DFKD methods and thus negatively affects the student training. To address this issue, we propose Customizing Synthetic Data for Data-Free Student Learning (CSD) in this paper, which achieves adaptive data synthesis using a self-supervised augmented auxiliary task to estimate the student learning ability. Specifically, data synthesis is dynamically adjusted to enlarge the cross entropy between the labels and the predictions from the self-supervised augmented task, thus generating hard samples for the student model. The experiments on various datasets and teacher-student models show the effectiveness of our proposed method. Code is available at: $\href{https://github.com/luoshiya/CSD}{https://github.com/luoshiya/CSD}$</abstract><doi>10.48550/arxiv.2307.04542</doi><oa>free_for_read</oa></addata></record>
fulltext fulltext_linktorsrc
identifier DOI: 10.48550/arxiv.2307.04542
ispartof
issn
language eng
recordid cdi_arxiv_primary_2307_04542
source arXiv.org
subjects Computer Science - Computer Vision and Pattern Recognition
title Customizing Synthetic Data for Data-Free Student Learning
url https://sfx.bib-bvb.de/sfx_tum?ctx_ver=Z39.88-2004&ctx_enc=info:ofi/enc:UTF-8&ctx_tim=2025-01-22T10%3A54%3A01IST&url_ver=Z39.88-2004&url_ctx_fmt=infofi/fmt:kev:mtx:ctx&rfr_id=info:sid/primo.exlibrisgroup.com:primo3-Article-arxiv_GOX&rft_val_fmt=info:ofi/fmt:kev:mtx:journal&rft.genre=article&rft.atitle=Customizing%20Synthetic%20Data%20for%20Data-Free%20Student%20Learning&rft.au=Luo,%20Shiya&rft.date=2023-07-10&rft_id=info:doi/10.48550/arxiv.2307.04542&rft_dat=%3Carxiv_GOX%3E2307_04542%3C/arxiv_GOX%3E%3Curl%3E%3C/url%3E&disable_directlink=true&sfx.directlink=off&sfx.report_link=0&rft_id=info:oai/&rft_id=info:pmid/&rfr_iscdi=true