Empirical study of privacy inference attack against deep reinforcement learning models
Most studies on privacy in machine learning have primarily focused on supervised learning, with little research on privacy concerns in reinforcement learning. However, our study has demonstrated that observation information can be extracted through trajectory analysis. In this paper, we propose a va...
Gespeichert in:
| Veröffentlicht in: | Connection science 2023-12, Vol.35 (1) |
|---|---|
| Hauptverfasser: | , , , |
| Format: | Artikel |
| Sprache: | eng |
| Schlagworte: | |
| Online-Zugang: | Volltext |
| Tags: |
Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
|
| container_end_page | |
|---|---|
| container_issue | 1 |
| container_start_page | |
| container_title | Connection science |
| container_volume | 35 |
| creator | Zhou, Huaicheng Mo, Kanghua Huang, Teng Li, Yongjin |
| description | Most studies on privacy in machine learning have primarily focused on supervised learning, with little research on privacy concerns in reinforcement learning. However, our study has demonstrated that observation information can be extracted through trajectory analysis. In this paper, we propose a variable information inference attack targeting the observation space of policy models, which is categorised into two types: observed value inference attack and observed variable inference. Our algorithm has demonstrated a high success rate in privacy inference attacks for both types of observation information. |
| doi_str_mv | 10.1080/09540091.2023.2211240 |
| format | Article |
| fullrecord | <record><control><sourceid>proquest_infor</sourceid><recordid>TN_cdi_proquest_journals_2907068951</recordid><sourceformat>XML</sourceformat><sourcesystem>PC</sourcesystem><sourcerecordid>2907068951</sourcerecordid><originalsourceid>FETCH-LOGICAL-c385t-42d68f429a73c288ce52bb5b24b86776b4bf52807fbe9e5ffc0086a0fb9e140e3</originalsourceid><addsrcrecordid>eNp9kMtOwzAQRS0EEqXwCUiWWKeMHTtxdqCqPKRKbICt5TjjKiVxip2C8vcktGxZzWLOvaM5hFwzWDBQcAuFFAAFW3Dg6YJzxriAEzJjaQYJiEKcktnEJBN0Ti5i3AKABMZm5H3V7upQW9PQ2O-rgXaO7kL9ZexAa-8woLdITd8b-0HNxtQ-9rRC3NGA474LFlv0PW3QBF_7DW27Cpt4Sc6caSJeHeecvD2sXpdPyfrl8Xl5v05sqmSfCF5lyglemDy1XCmLkpelLLkoVZbnWSlKJ7mC3JVYoHTOAqjMgCsLZAIwnZObQ-8udJ97jL3edvvgx5OaF5BDpgrJRkoeKBu6GAM6Pb7YmjBoBnpSqP8U6kmhPiocc3eH3O-nrfnuQlPp3gxNF1ww3tZRp_9X_ADr5XiU</addsrcrecordid><sourcetype>Aggregation Database</sourcetype><iscdi>true</iscdi><recordtype>article</recordtype><pqid>2907068951</pqid></control><display><type>article</type><title>Empirical study of privacy inference attack against deep reinforcement learning models</title><source>Taylor & Francis Open Access</source><source>Alma/SFX Local Collection</source><creator>Zhou, Huaicheng ; Mo, Kanghua ; Huang, Teng ; Li, Yongjin</creator><creatorcontrib>Zhou, Huaicheng ; Mo, Kanghua ; Huang, Teng ; Li, Yongjin</creatorcontrib><description>Most studies on privacy in machine learning have primarily focused on supervised learning, with little research on privacy concerns in reinforcement learning. However, our study has demonstrated that observation information can be extracted through trajectory analysis. In this paper, we propose a variable information inference attack targeting the observation space of policy models, which is categorised into two types: observed value inference attack and observed variable inference. Our algorithm has demonstrated a high success rate in privacy inference attacks for both types of observation information.</description><identifier>ISSN: 0954-0091</identifier><identifier>EISSN: 1360-0494</identifier><identifier>DOI: 10.1080/09540091.2023.2211240</identifier><language>eng</language><publisher>Abingdon: Taylor & Francis</publisher><subject>adversarial attack ; Algorithms ; data security ; Deep learning ; Empirical analysis ; Inference ; Machine learning ; Privacy ; privacy attack ; Reinforcement learning ; Supervised learning ; Trajectory analysis</subject><ispartof>Connection science, 2023-12, Vol.35 (1)</ispartof><rights>2023 The Author(s). Published by Informa UK Limited, trading as Taylor & Francis Group. 2023</rights><rights>2023 The Author(s). Published by Informa UK Limited, trading as Taylor & Francis Group. This work is licensed under the Creative Commons Attribution – Non-Commercial License http://creativecommons.org/licenses/by-nc/4.0/ (the “License”). Notwithstanding the ProQuest Terms and Conditions, you may use this content in accordance with the terms of the License.</rights><lds50>peer_reviewed</lds50><oa>free_for_read</oa><woscitedreferencessubscribed>false</woscitedreferencessubscribed><citedby>FETCH-LOGICAL-c385t-42d68f429a73c288ce52bb5b24b86776b4bf52807fbe9e5ffc0086a0fb9e140e3</citedby><cites>FETCH-LOGICAL-c385t-42d68f429a73c288ce52bb5b24b86776b4bf52807fbe9e5ffc0086a0fb9e140e3</cites></display><links><openurl>$$Topenurl_article</openurl><openurlfulltext>$$Topenurlfull_article</openurlfulltext><thumbnail>$$Tsyndetics_thumb_exl</thumbnail><linktopdf>$$Uhttps://www.tandfonline.com/doi/pdf/10.1080/09540091.2023.2211240$$EPDF$$P50$$Ginformaworld$$Hfree_for_read</linktopdf><linktohtml>$$Uhttps://www.tandfonline.com/doi/full/10.1080/09540091.2023.2211240$$EHTML$$P50$$Ginformaworld$$Hfree_for_read</linktohtml><link.rule.ids>314,776,780,27479,27901,27902,59116,59117</link.rule.ids></links><search><creatorcontrib>Zhou, Huaicheng</creatorcontrib><creatorcontrib>Mo, Kanghua</creatorcontrib><creatorcontrib>Huang, Teng</creatorcontrib><creatorcontrib>Li, Yongjin</creatorcontrib><title>Empirical study of privacy inference attack against deep reinforcement learning models</title><title>Connection science</title><description>Most studies on privacy in machine learning have primarily focused on supervised learning, with little research on privacy concerns in reinforcement learning. However, our study has demonstrated that observation information can be extracted through trajectory analysis. In this paper, we propose a variable information inference attack targeting the observation space of policy models, which is categorised into two types: observed value inference attack and observed variable inference. Our algorithm has demonstrated a high success rate in privacy inference attacks for both types of observation information.</description><subject>adversarial attack</subject><subject>Algorithms</subject><subject>data security</subject><subject>Deep learning</subject><subject>Empirical analysis</subject><subject>Inference</subject><subject>Machine learning</subject><subject>Privacy</subject><subject>privacy attack</subject><subject>Reinforcement learning</subject><subject>Supervised learning</subject><subject>Trajectory analysis</subject><issn>0954-0091</issn><issn>1360-0494</issn><fulltext>true</fulltext><rsrctype>article</rsrctype><creationdate>2023</creationdate><recordtype>article</recordtype><sourceid>0YH</sourceid><recordid>eNp9kMtOwzAQRS0EEqXwCUiWWKeMHTtxdqCqPKRKbICt5TjjKiVxip2C8vcktGxZzWLOvaM5hFwzWDBQcAuFFAAFW3Dg6YJzxriAEzJjaQYJiEKcktnEJBN0Ti5i3AKABMZm5H3V7upQW9PQ2O-rgXaO7kL9ZexAa-8woLdITd8b-0HNxtQ-9rRC3NGA474LFlv0PW3QBF_7DW27Cpt4Sc6caSJeHeecvD2sXpdPyfrl8Xl5v05sqmSfCF5lyglemDy1XCmLkpelLLkoVZbnWSlKJ7mC3JVYoHTOAqjMgCsLZAIwnZObQ-8udJ97jL3edvvgx5OaF5BDpgrJRkoeKBu6GAM6Pb7YmjBoBnpSqP8U6kmhPiocc3eH3O-nrfnuQlPp3gxNF1ww3tZRp_9X_ADr5XiU</recordid><startdate>20231231</startdate><enddate>20231231</enddate><creator>Zhou, Huaicheng</creator><creator>Mo, Kanghua</creator><creator>Huang, Teng</creator><creator>Li, Yongjin</creator><general>Taylor & Francis</general><general>Taylor & Francis Ltd</general><scope>0YH</scope><scope>AAYXX</scope><scope>CITATION</scope><scope>JQ2</scope><scope>NAPCQ</scope></search><sort><creationdate>20231231</creationdate><title>Empirical study of privacy inference attack against deep reinforcement learning models</title><author>Zhou, Huaicheng ; Mo, Kanghua ; Huang, Teng ; Li, Yongjin</author></sort><facets><frbrtype>5</frbrtype><frbrgroupid>cdi_FETCH-LOGICAL-c385t-42d68f429a73c288ce52bb5b24b86776b4bf52807fbe9e5ffc0086a0fb9e140e3</frbrgroupid><rsrctype>articles</rsrctype><prefilter>articles</prefilter><language>eng</language><creationdate>2023</creationdate><topic>adversarial attack</topic><topic>Algorithms</topic><topic>data security</topic><topic>Deep learning</topic><topic>Empirical analysis</topic><topic>Inference</topic><topic>Machine learning</topic><topic>Privacy</topic><topic>privacy attack</topic><topic>Reinforcement learning</topic><topic>Supervised learning</topic><topic>Trajectory analysis</topic><toplevel>peer_reviewed</toplevel><toplevel>online_resources</toplevel><creatorcontrib>Zhou, Huaicheng</creatorcontrib><creatorcontrib>Mo, Kanghua</creatorcontrib><creatorcontrib>Huang, Teng</creatorcontrib><creatorcontrib>Li, Yongjin</creatorcontrib><collection>Taylor & Francis Open Access</collection><collection>CrossRef</collection><collection>ProQuest Computer Science Collection</collection><collection>Nursing & Allied Health Premium</collection><jtitle>Connection science</jtitle></facets><delivery><delcategory>Remote Search Resource</delcategory><fulltext>fulltext</fulltext></delivery><addata><au>Zhou, Huaicheng</au><au>Mo, Kanghua</au><au>Huang, Teng</au><au>Li, Yongjin</au><format>journal</format><genre>article</genre><ristype>JOUR</ristype><atitle>Empirical study of privacy inference attack against deep reinforcement learning models</atitle><jtitle>Connection science</jtitle><date>2023-12-31</date><risdate>2023</risdate><volume>35</volume><issue>1</issue><issn>0954-0091</issn><eissn>1360-0494</eissn><abstract>Most studies on privacy in machine learning have primarily focused on supervised learning, with little research on privacy concerns in reinforcement learning. However, our study has demonstrated that observation information can be extracted through trajectory analysis. In this paper, we propose a variable information inference attack targeting the observation space of policy models, which is categorised into two types: observed value inference attack and observed variable inference. Our algorithm has demonstrated a high success rate in privacy inference attacks for both types of observation information.</abstract><cop>Abingdon</cop><pub>Taylor & Francis</pub><doi>10.1080/09540091.2023.2211240</doi><oa>free_for_read</oa></addata></record> |
| fulltext | fulltext |
| identifier | ISSN: 0954-0091 |
| ispartof | Connection science, 2023-12, Vol.35 (1) |
| issn | 0954-0091 1360-0494 |
| language | eng |
| recordid | cdi_proquest_journals_2907068951 |
| source | Taylor & Francis Open Access; Alma/SFX Local Collection |
| subjects | adversarial attack Algorithms data security Deep learning Empirical analysis Inference Machine learning Privacy privacy attack Reinforcement learning Supervised learning Trajectory analysis |
| title | Empirical study of privacy inference attack against deep reinforcement learning models |
| url | https://sfx.bib-bvb.de/sfx_tum?ctx_ver=Z39.88-2004&ctx_enc=info:ofi/enc:UTF-8&ctx_tim=2025-02-11T20%3A21%3A18IST&url_ver=Z39.88-2004&url_ctx_fmt=infofi/fmt:kev:mtx:ctx&rfr_id=info:sid/primo.exlibrisgroup.com:primo3-Article-proquest_infor&rft_val_fmt=info:ofi/fmt:kev:mtx:journal&rft.genre=article&rft.atitle=Empirical%20study%20of%20privacy%20inference%20attack%20against%20deep%20reinforcement%20learning%20models&rft.jtitle=Connection%20science&rft.au=Zhou,%20Huaicheng&rft.date=2023-12-31&rft.volume=35&rft.issue=1&rft.issn=0954-0091&rft.eissn=1360-0494&rft_id=info:doi/10.1080/09540091.2023.2211240&rft_dat=%3Cproquest_infor%3E2907068951%3C/proquest_infor%3E%3Curl%3E%3C/url%3E&disable_directlink=true&sfx.directlink=off&sfx.report_link=0&rft_id=info:oai/&rft_pqid=2907068951&rft_id=info:pmid/&rfr_iscdi=true |