講演名 2020-01-17
[Poster Presentation] Investigation of neuron with sigmoid activation function for quantum-flux-parametron-based neural network
山口 大貴(横浜国大), 山梨 裕希(横浜国大), 吉川 信行(横浜国大),
PDFダウンロードページ PDFダウンロードページへ
抄録(和) An ANN is a circuit that imitates the brain of a human, which is effective for pattern recognition and ?ltering, and so on. A superconducting circuit can operate at the high fre-quency with low power consumption, it is attractive for realization of energy-ef?cient ANNs. In order to perform ANN learning, it is necessary for the activation function of neurons in the middle layer to have characteristics like sigmoid/hyperbolic tangent function. The rf-superconducting quantum interference device (rf-SQUID) is composed of a supercon-ducting loop including a Josephson junction and an inductance, and when a magnetic ?ux is applied to the circuit, a circular current is generated as if sine is inclined. Applying the cir-cuit of rf-SQUID, it is known that the input/output relationship of magnetic ?ux has charac-teristics close to the sigmoid/hyperbolic tangent function. In this study, we designed a neuron with a hyperbolic tangent activation function using the AIST fabrication process. Fur-thermore, we investigated tuning method of the quantum-flux-parametron-based supercon-ducting neural network toward its learning.
抄録(英) An ANN is a circuit that imitates the brain of a human, which is effective for pattern recognition and ?ltering, and so on. A superconducting circuit can operate at the high fre-quency with low power consumption, it is attractive for realization of energy-ef?cient ANNs. In order to perform ANN learning, it is necessary for the activation function of neurons in the middle layer to have characteristics like sigmoid/hyperbolic tangent function. The rf-superconducting quantum interference device (rf-SQUID) is composed of a supercon-ducting loop including a Josephson junction and an inductance, and when a magnetic ?ux is applied to the circuit, a circular current is generated as if sine is inclined. Applying the cir-cuit of rf-SQUID, it is known that the input/output relationship of magnetic ?ux has charac-teristics close to the sigmoid/hyperbolic tangent function. In this study, we designed a neuron with a hyperbolic tangent activation function using the AIST fabrication process. Fur-thermore, we investigated tuning method of the quantum-flux-parametron-based supercon-ducting neural network toward its learning.
キーワード(和)
キーワード(英) rf-SQUIDANNsigmoidhyperbolic tangent
資料番号 SCE2019-60
発行日 2020-01-09 (SCE)

研究会情報
研究会 SCE
開催期間 2020/1/16(から2日開催)
開催地(和) 横浜市開港記念会館
開催地(英)
テーマ(和) 超伝導エレクトロニクス一般
テーマ(英)
委員長氏名(和) 神代 暁(産総研)
委員長氏名(英) Satoshi Kohjiro(AIST)
副委員長氏名(和)
副委員長氏名(英)
幹事氏名(和) 竹内 尚輝(横浜国大) / 三木 茂人(NICT)
幹事氏名(英) Naoki Takeuchi(Yokohama National Univ.) / Shigehito Miki(NICT)
幹事補佐氏名(和) 赤池 宏之(大同大)
幹事補佐氏名(英) Hiroyuki Akaike(Daido Univ.)

講演論文情報詳細
申込み研究会 Technical Committee on Superconductive Electronics
本文の言語 ENG
タイトル(和)
サブタイトル(和)
タイトル(英) [Poster Presentation] Investigation of neuron with sigmoid activation function for quantum-flux-parametron-based neural network
サブタイトル(和)
キーワード(1)(和/英) / rf-SQUIDANNsigmoidhyperbolic tangent
第 1 著者 氏名(和/英) 山口 大貴 / Daiki Yamaguchi
第 1 著者 所属(和/英) 横浜国立大学(略称:横浜国大)
Yokohama National University(略称:Yokohama Natl. Univ.)
第 2 著者 氏名(和/英) 山梨 裕希 / Yuki Yamanashi
第 2 著者 所属(和/英) 横浜国立大学(略称:横浜国大)
Yokohama National University(略称:Yokohama Natl. Univ.)
第 3 著者 氏名(和/英) 吉川 信行 / Nobuyuki Yoshikawa
第 3 著者 所属(和/英) 横浜国立大学(略称:横浜国大)
Yokohama National University(略称:Yokohama Natl. Univ.)
発表年月日 2020-01-17
資料番号 SCE2019-60
巻番号(vol) vol.119
号番号(no) SCE-369
ページ範囲 pp.125-127(SCE),
ページ数 3
発行日 2020-01-09 (SCE)