WEKO3
インデックスリンク
アイテム
{"_buckets": {"deposit": "3896a629-ad68-47f0-a125-a1cbef240613"}, "_deposit": {"created_by": 3, "id": "7926", "owners": [3], "pid": {"revision_id": 0, "type": "depid", "value": "7926"}, "status": "published"}, "_oai": {"id": "oai:kanazawa-u.repo.nii.ac.jp:00007926", "sets": ["936"]}, "author_link": ["353", "10909"], "item_8_biblio_info_8": {"attribute_name": "書誌情報", "attribute_value_mlt": [{"bibliographicIssueDates": {"bibliographicIssueDate": "1994-01-01", "bibliographicIssueDateType": "Issued"}, "bibliographicPageEnd": "3002", "bibliographicPageStart": "2997", "bibliographicVolumeNumber": "5", "bibliographic_titles": [{"bibliographic_title": "IEEE International Conference on Neural Networks - Conference Proceedings"}]}]}, "item_8_description_21": {"attribute_name": "抄録", "attribute_value_mlt": [{"subitem_description": "This paper discusses properties of activation functions in multilayer neural network applied to pattern classification. A rule of thumb for selecting activation functions or their combination is proposed. The sigmoid, Gaussian and sinusoidal functions are selected due to their independent and fundamental space division properties. The sigmoid function is not effective for a single hidden unit. On the contrary, the other functions can provide good performance. When several hidden units are employed, the sigmoid function is useful. However, the convergence speed is still slower than the others. The Gaussian function is sensitive to the additive noise, while the others are rather insensitive. As a result, based on convergence rates, the minimum error and noise sensitivity, the sinusoidal function is most useful for both without and with additive noise. Property of each function is discussed based on the internal representation, that is the distributions of the hidden unit inputs and outputs. Although this selection depends on the input signals to be classified, the periodic function can be effectively applied to a wide range of application fields.", "subitem_description_type": "Abstract"}]}, "item_8_publisher_17": {"attribute_name": "出版者", "attribute_value_mlt": [{"subitem_publisher": "Institute of Electrical and Electronics Engineers (IEEE)"}]}, "item_8_source_id_9": {"attribute_name": "ISSN", "attribute_value_mlt": [{"subitem_source_identifier": "1098-7576", "subitem_source_identifier_type": "ISSN"}]}, "item_8_version_type_25": {"attribute_name": "著者版フラグ", "attribute_value_mlt": [{"subitem_version_resource": "http://purl.org/coar/version/c_970fb48d4fbd8a85", "subitem_version_type": "VoR"}]}, "item_creator": {"attribute_name": "著者", "attribute_type": "creator", "attribute_value_mlt": [{"creatorNames": [{"creatorName": "Hara, Kazuyuki"}], "nameIdentifiers": [{"nameIdentifier": "10909", "nameIdentifierScheme": "WEKO"}]}, {"creatorNames": [{"creatorName": "Nakayama, Kenji"}], "nameIdentifiers": [{"nameIdentifier": "353", "nameIdentifierScheme": "WEKO"}, {"nameIdentifier": "00207945", "nameIdentifierScheme": "e-Rad", "nameIdentifierURI": "https://kaken.nii.ac.jp/ja/search/?qm=00207945"}, {"nameIdentifier": "00207945", "nameIdentifierScheme": "研究者番号", "nameIdentifierURI": "https://nrid.nii.ac.jp/nrid/1000000207945"}]}]}, "item_files": {"attribute_name": "ファイル情報", "attribute_type": "file", "attribute_value_mlt": [{"accessrole": "open_date", "date": [{"dateType": "Available", "dateValue": "2017-10-03"}], "displaytype": "detail", "download_preview_message": "", "file_order": 0, "filename": "TE-PR-NAKAYAMA-K-2997.pdf", "filesize": [{"value": "666.7 kB"}], "format": "application/pdf", "future_date_message": "", "is_thumbnail": false, "licensetype": "license_free", "mimetype": "application/pdf", "size": 666700.0, "url": {"label": "TE-PR-NAKAYAMA-K-2997.pdf", "url": "https://kanazawa-u.repo.nii.ac.jp/record/7926/files/TE-PR-NAKAYAMA-K-2997.pdf"}, "version_id": "7eec73b5-6b4a-4689-bbf0-5e4ef63c01d7"}]}, "item_language": {"attribute_name": "言語", "attribute_value_mlt": [{"subitem_language": "eng"}]}, "item_resource_type": {"attribute_name": "資源タイプ", "attribute_value_mlt": [{"resourcetype": "conference paper", "resourceuri": "http://purl.org/coar/resource_type/c_5794"}]}, "item_title": "Comparison of activation functions in multilayer neural network for pattern classification", "item_titles": {"attribute_name": "タイトル", "attribute_value_mlt": [{"subitem_title": "Comparison of activation functions in multilayer neural network for pattern classification"}]}, "item_type_id": "8", "owner": "3", "path": ["936"], "permalink_uri": "http://hdl.handle.net/2297/11893", "pubdate": {"attribute_name": "公開日", "attribute_value": "2017-10-03"}, "publish_date": "2017-10-03", "publish_status": "0", "recid": "7926", "relation": {}, "relation_version_is_last": true, "title": ["Comparison of activation functions in multilayer neural network for pattern classification"], "weko_shared_id": -1}
Comparison of activation functions in multilayer neural network for pattern classification
http://hdl.handle.net/2297/11893
http://hdl.handle.net/2297/11893c18f5cf3-c3eb-4a37-9977-edce43361598
名前 / ファイル | ライセンス | アクション |
---|---|---|
TE-PR-NAKAYAMA-K-2997.pdf (666.7 kB)
|
|
Item type | 会議発表論文 / Conference Paper(1) | |||||
---|---|---|---|---|---|---|
公開日 | 2017-10-03 | |||||
タイトル | ||||||
タイトル | Comparison of activation functions in multilayer neural network for pattern classification | |||||
言語 | ||||||
言語 | eng | |||||
資源タイプ | ||||||
資源タイプ識別子 | http://purl.org/coar/resource_type/c_5794 | |||||
資源タイプ | conference paper | |||||
著者 |
Hara, Kazuyuki
× Hara, Kazuyuki× Nakayama, Kenji |
|||||
書誌情報 |
IEEE International Conference on Neural Networks - Conference Proceedings 巻 5, p. 2997-3002, 発行日 1994-01-01 |
|||||
ISSN | ||||||
収録物識別子タイプ | ISSN | |||||
収録物識別子 | 1098-7576 | |||||
出版者 | ||||||
出版者 | Institute of Electrical and Electronics Engineers (IEEE) | |||||
抄録 | ||||||
内容記述タイプ | Abstract | |||||
内容記述 | This paper discusses properties of activation functions in multilayer neural network applied to pattern classification. A rule of thumb for selecting activation functions or their combination is proposed. The sigmoid, Gaussian and sinusoidal functions are selected due to their independent and fundamental space division properties. The sigmoid function is not effective for a single hidden unit. On the contrary, the other functions can provide good performance. When several hidden units are employed, the sigmoid function is useful. However, the convergence speed is still slower than the others. The Gaussian function is sensitive to the additive noise, while the others are rather insensitive. As a result, based on convergence rates, the minimum error and noise sensitivity, the sinusoidal function is most useful for both without and with additive noise. Property of each function is discussed based on the internal representation, that is the distributions of the hidden unit inputs and outputs. Although this selection depends on the input signals to be classified, the periodic function can be effectively applied to a wide range of application fields. | |||||
著者版フラグ | ||||||
出版タイプ | VoR | |||||
出版タイプResource | http://purl.org/coar/version/c_970fb48d4fbd8a85 |