High-Accuracy and Cost-Effective Neural Networks for Embedded Systems

• Item type: SIG Technical Reports(1)
• 公開日: 2019-03-10
• タイトル: High-Accuracy and Cost-Effective Neural Networks for Embedded Systems
• タイトル / 言語: en
• タイトル: High-Accuracy and Cost-Effective Neural Networks for Embedded Systems
• 言語: eng
• キーワード / 主題Scheme: Other
• キーワード / 主題: 機械学習/深層学習
• 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: Japan Advanced Institute of Science and Technology JAIST
• 著者所属: Japan Advanced Institute of Science and Technology JAIST
• 著者所属: Japan Advanced Institute of Science and Technology JAIST
• 著者所属: Japan Advanced Institute of Science and Technology JAIST
• 著者所属(英): en
• 著者所属(英): Japan Advanced Institute of Science and Technology JAIST
• 著者所属(英): en
• 著者所属(英): Japan Advanced Institute of Science and Technology JAIST
• 著者所属(英): en
• 著者所属(英): Japan Advanced Institute of Science and Technology JAIST
• 著者所属(英): en
• 著者所属(英): Japan Advanced Institute of Science and Technology JAIST
• 著者名: Jiajun, Guo; Amr, Ashmawy; Thiem, Van Chu; Kiyofumi, Tanaka
• 著者名(英): Jiajun, Guo; Amr, Ashmawy; Thiem, Van Chu; Kiyofumi, Tanaka
• 論文抄録 / 内容記述タイプ: Other
• 論文抄録 / 内容記述: Binarized Neural Network (BNN) is a promising technique for embedded inference hardware due to the small hardware cost, but the inference accuracy is degraded compared to a full-precision CNN. In this study, we show that the inference accuracy can be improved by using an ensemble of a few BNNs. In addition, we report our implementation on an FPGA.
• 論文抄録(英) / 内容記述タイプ: Other
• 論文抄録(英) / 内容記述: Binarized Neural Network (BNN) is a promising technique for embedded inference hardware due to the small hardware cost, but the inference accuracy is degraded compared to a full-precision CNN. In this study, we show that the inference accuracy can be improved by using an ensemble of a few BNNs. In addition, we report our implementation on an FPGA.
• 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: AN10096105
• 書誌情報: 研究報告システム・アーキテクチャ（ARC）; 巻 2019-ARC-235, 号 36, p. 1-8, 発行日 2019-03-10
• ISSN / 収録物識別子タイプ: ISSN
• ISSN / 収録物識別子: 2188-8574
• 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