Computational Cost Reduction for Human Activity Recognition Using Clockwork Recurrent Neural Networks

• Item type: SIG Technical Reports(1)
• 公開日: 2022-08-29
• タイトル: Computational Cost Reduction for Human Activity Recognition Using Clockwork Recurrent Neural Networks
• タイトル / 言語: en
• タイトル: Computational Cost Reduction for Human Activity Recognition Using Clockwork Recurrent Neural Networks
• 言語: eng
• キーワード / 主題Scheme: Other
• キーワード / 主題: 生体センシング
• 資源タイプ / 資源タイプ識別子: http://purl.org/coar/resource_type/c_18gh
• 資源タイプ: technical report
• 著者所属: 国際基督教大学
• 著者所属: 国際基督教大学
• 著者所属(英): en
• 著者所属(英): International Christian University
• 著者所属(英): en
• 著者所属(英): International Christian University
• 著者名: 山岸, 信夫; 鏑木, 崇史
• 著者名(英): Shinobu, Yamagishi; Takashi, Kaburagi
• 論文抄録 / 内容記述タイプ: Other
• 論文抄録 / 内容記述: Human Activity Recognition is a field concerned with the identification of human activities using digital sensors such as cameras and motion sensors. The field is becoming increasingly relevant to the daily lives of people in areas with an abundance of technology in the form of “smart” devices. The smartphone is an popular example of a device that can be used to detect activities. However, the current methods for identifying activities tend to have a high computational cost that prevent them from occuring on devices with less computation capabilities. Thus, in this paper we suggest the Clockwork Recurrent Neural Network (CWRNN) model developed by Koutnik in 2014 [1] as a way to decrease computational cost. The results of the implementation of CWRNN against Long Short-Term Memory (LSTM) suggest that LSTMs are superior for classification accuracy, though the CWRNN model used has potential to be further optimized
• 論文抄録(英) / 内容記述タイプ: Other
• 論文抄録(英) / 内容記述: Human Activity Recognition is a field concerned with the identification of human activities using digital sensors such as cameras and motion sensors. The field is becoming increasingly relevant to the daily lives of people in areas with an abundance of technology in the form of “smart” devices. The smartphone is an popular example of a device that can be used to detect activities. However, the current methods for identifying activities tend to have a high computational cost that prevent them from occuring on devices with less computation capabilities. Thus, in this paper we suggest the Clockwork Recurrent Neural Network (CWRNN) model developed by Koutnik in 2014 [1] as a way to decrease computational cost. The results of the implementation of CWRNN against Long Short-Term Memory (LSTM) suggest that LSTMs are superior for classification accuracy, though the CWRNN model used has potential to be further optimized
• 書誌レコードID / 収録物識別子タイプ: NCID
• 書誌レコードID / 収録物識別子: AA12628327
• 書誌情報: 研究報告コンシューマ・デバイス＆システム（CDS）; 巻 2022-CDS-35, 号 1, p. 1-7, 発行日 2022-08-29
• ISSN / 収録物識別子タイプ: ISSN
• ISSN / 収録物識別子: 2188-8604
• Notice: SIG Technical Reports are nonrefereed and hence may later appear in any journals, conferences, symposia, etc.
• 出版者 / 言語: ja
• 出版者: 情報処理学会