{"created":"2025-01-19T01:37:08.828804+00:00","updated":"2025-01-19T09:36:02.838047+00:00","metadata":{"_oai":{"id":"oai:ipsj.ixsq.nii.ac.jp:00235315","sets":["1164:3782:11460:11657"]},"path":["11657"],"owner":"44499","recid":"235315","title":["大規模言語モデルと弱教師付き学習を利用したゼロショット分類:スマートホームにおける行動認識への適用"],"pubdate":{"attribute_name":"公開日","attribute_value":"2024-07-04"},"_buckets":{"deposit":"58ff3d92-68cd-4523-be8d-8d3ed622a313"},"_deposit":{"id":"235315","pid":{"type":"depid","value":"235315","revision_id":0},"owners":[44499],"status":"published","created_by":44499},"item_title":"大規模言語モデルと弱教師付き学習を利用したゼロショット分類:スマートホームにおける行動認識への適用","author_link":["642818","642822","642824","642820","642817","642823","642819","642821"],"item_titles":{"attribute_name":"タイトル","attribute_value_mlt":[{"subitem_title":"大規模言語モデルと弱教師付き学習を利用したゼロショット分類:スマートホームにおける行動認識への適用"},{"subitem_title":"Zero-Shot Classification Using Large Language Models and Weak Supervision: Application of Human Activity Recognition in Smart Home","subitem_title_language":"en"}]},"item_keyword":{"attribute_name":"キーワード","attribute_value_mlt":[{"subitem_subject":"LOIS 4","subitem_subject_scheme":"Other"}]},"item_type_id":"4","publish_date":"2024-07-04","item_4_text_3":{"attribute_name":"著者所属","attribute_value_mlt":[{"subitem_text_value":"KDDI総合研究所"},{"subitem_text_value":"KDDI総合研究所"},{"subitem_text_value":"KDDI総合研究所"},{"subitem_text_value":"KDDI総合研究所"}]},"item_4_text_4":{"attribute_name":"著者所属(英)","attribute_value_mlt":[{"subitem_text_value":"KDDI Research, Inc.","subitem_text_language":"en"},{"subitem_text_value":"KDDI Research, Inc.","subitem_text_language":"en"},{"subitem_text_value":"KDDI Research, Inc.","subitem_text_language":"en"},{"subitem_text_value":"KDDI Research, Inc.","subitem_text_language":"en"}]},"item_language":{"attribute_name":"言語","attribute_value_mlt":[{"subitem_language":"jpn"}]},"item_publisher":{"attribute_name":"出版者","attribute_value_mlt":[{"subitem_publisher":"情報処理学会","subitem_publisher_language":"ja"}]},"publish_status":"0","weko_shared_id":-1,"item_file_price":{"attribute_name":"Billing file","attribute_type":"file","attribute_value_mlt":[{"url":{"url":"https://ipsj.ixsq.nii.ac.jp/record/235315/files/IPSJ-DC24133014.pdf","label":"IPSJ-DC24133014.pdf"},"format":"application/pdf","billing":["billing_file"],"filename":"IPSJ-DC24133014.pdf","filesize":[{"value":"1.2 MB"}],"mimetype":"application/pdf","priceinfo":[{"tax":["include_tax"],"price":"0","billingrole":"32"},{"tax":["include_tax"],"price":"0","billingrole":"44"}],"accessrole":"open_login","version_id":"397883e8-dfea-41c0-9b62-cd0442452088","displaytype":"detail","licensetype":"license_note","license_note":"Copyright (c) 2024 by the Institute of Electronics, Information and CommunicationEngineers This SIG report is only available to those in membership of the SIG."}]},"item_4_creator_5":{"attribute_name":"著者名","attribute_type":"creator","attribute_value_mlt":[{"creatorNames":[{"creatorName":"武田, 直人"}],"nameIdentifiers":[{}]},{"creatorNames":[{"creatorName":"西村, 康孝"}],"nameIdentifiers":[{}]},{"creatorNames":[{"creatorName":"山崎, 悠大"}],"nameIdentifiers":[{}]},{"creatorNames":[{"creatorName":"池田, 和史"}],"nameIdentifiers":[{}]}]},"item_4_creator_6":{"attribute_name":"著者名(英)","attribute_type":"creator","attribute_value_mlt":[{"creatorNames":[{"creatorName":"Naoto, Takeda","creatorNameLang":"en"}],"nameIdentifiers":[{}]},{"creatorNames":[{"creatorName":"Yasutaka, Nishimura","creatorNameLang":"en"}],"nameIdentifiers":[{}]},{"creatorNames":[{"creatorName":"Yudai, Yamazaki","creatorNameLang":"en"}],"nameIdentifiers":[{}]},{"creatorNames":[{"creatorName":"Kazushi, Ikeda","creatorNameLang":"en"}],"nameIdentifiers":[{}]}]},"item_4_source_id_9":{"attribute_name":"書誌レコードID","attribute_value_mlt":[{"subitem_source_identifier":"AN10539261","subitem_source_identifier_type":"NCID"}]},"item_4_textarea_12":{"attribute_name":"Notice","attribute_value_mlt":[{"subitem_textarea_value":"SIG Technical Reports are nonrefereed and hence may later appear in any journals, conferences, symposia, etc."}]},"item_resource_type":{"attribute_name":"資源タイプ","attribute_value_mlt":[{"resourceuri":"http://purl.org/coar/resource_type/c_18gh","resourcetype":"technical report"}]},"item_4_source_id_11":{"attribute_name":"ISSN","attribute_value_mlt":[{"subitem_source_identifier":"2188-8892","subitem_source_identifier_type":"ISSN"}]},"item_4_description_7":{"attribute_name":"論文抄録","attribute_value_mlt":[{"subitem_description":"スマートホームに設置した環境センサを用いた居住者の行動認識により,健康モニタリングや異常検知等のサービスを提供できる.居住者やスマートホームごとに生活習慣やセンサ種別が異なるため,行動認識モデルの構築には,居住者本人によるラベリングが必要だが負担が大きい.GPT-4 や Gemini 等の大規模言語モデル(Large Language Models: LLMs)を活用することで,居住者にラベリングを依頼することなく,入居の初期段階からサービスを提供できる可能性がある.本研究では,LLM を用いた行動認識のゼロショット性能を向上させる手法を提案する.具体的には,LLM からゼロショット推定結果だけでなく,推定根拠となったセンサ群の情報を収集し,分類のためのラベリング関数群を自動構築する.さらにラベリング関数に含まれるノイズの影響を低減するために弱教師付き学習の枠組みを利用し推定結果を統合する.7 名の実験参加者によるデータセットを用いた実験の結果,ゼロショット性能よりも平均で 0.40 から 0.51 へと 0.11 の F1 値の向上が確認された.この性能の達成には,従来の方法では,約 40 日間にわたる居住者によるラベリングが必要であった.","subitem_description_type":"Other"}]},"item_4_description_8":{"attribute_name":"論文抄録(英)","attribute_value_mlt":[{"subitem_description":"Sensor-based activity recognition in smart homes enables applications like health monitoring andanomaly detection. Due to the varying lifestyles of residents and the different sensor types in smart homes, self-labeling is needed to build an activity recognition model, which can be burdensome. By leveraging LLMs such as GPT-4 and Gemini, smart home services can be offered from the start without labeled data. We propose a method to enhance zero-shot activity recognition using LLM. This involves collecting zero-shot estimations and the sensor information underlying them to automatically create labeling functions for classification. We integrate the estimation results using a weak supervision framework to reduce noise. Experiments with data from seven participants showed an average F1 score improvement of 0.11 from 0.40 to 0.51 compared to zero-shot performance. Achieving this performance with traditional methods, which require labeling by the residents, would necessitate about 40 days.","subitem_description_type":"Other"}]},"item_4_biblio_info_10":{"attribute_name":"書誌情報","attribute_value_mlt":[{"bibliographicPageEnd":"6","bibliographic_titles":[{"bibliographic_title":"研究報告ドキュメントコミュニケーション(DC)"}],"bibliographicPageStart":"1","bibliographicIssueDates":{"bibliographicIssueDate":"2024-07-04","bibliographicIssueDateType":"Issued"},"bibliographicIssueNumber":"14","bibliographicVolumeNumber":"2024-DC-133"}]},"relation_version_is_last":true,"weko_creator_id":"44499"},"id":235315,"links":{}}