Learning method for a quantum bit network

Yoshihiro Osakabe, Shigeo Sato, Mitsunaga Kinjo, Koji Nakajima, Hisanao Akima, Masao Sakuraba

研究成果: Conference contribution

2 被引用数 (Scopus)

抄録

Quantum computing (QC) has attracted much attention due to its enormous computing power, but proposed algorithms so far are not sufficient for practical use. Therefore, if a quantum computer could obtain algorithms by itself, the applicable field of QC would be extended greatly. In this study, we investigate a learning method for a quantum bit network (QBN) by utilizing the analogy between an artificial neural network and a QBN as described in the previous reports [1, 2]. According to this analogy, we can relate a synaptic weight matrix with a Hamiltonian. We propose a quantum version of Hebb learning as follows; we enhance both excitatory and inhibitory couplings according to the probability that arbitrary two quantum bits (qubits) take the same or opposite states when a QBN outputs a desired pattern. As a first step, we trained a QBN shown in Fig. 1 to learn the XOR problem. We updated the Hamiltonian only when the hidden qubit took the state “1” in order to break symmetry because the network always learns a pair of symmetric patterns whether these patterns are desired or not. A typical successful learning result is shown in Fig. 2. Though the success rate of learning with various initial Hamiltonians reaches only 50%, this preliminary result indicates certain possibility for implementing learning function with a QBN.

本文言語English
ホスト出版物のタイトルArtificial Neural Networks and Machine Learning - 25th International Conference on Artificial Neural Networks, ICANN 2016, Proceedings
編集者Alessandro E.P. Villa, Paolo Masulli, Antonio Javier Pons Rivero
出版社Springer Verlag
ページ558-559
ページ数2
ISBN(印刷版)9783319447773
出版ステータスPublished - 2016
イベント25th International Conference on Artificial Neural Networks, ICANN 2016 - Barcelona, Spain
継続期間: 2016 9 62016 9 9

出版物シリーズ

名前Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
9886 LNCS
ISSN(印刷版)0302-9743
ISSN(電子版)1611-3349

Other

Other25th International Conference on Artificial Neural Networks, ICANN 2016
国/地域Spain
CityBarcelona
Period16/9/616/9/9

ASJC Scopus subject areas

  • 理論的コンピュータサイエンス
  • コンピュータ サイエンス(全般)

フィンガープリント

「Learning method for a quantum bit network」の研究トピックを掘り下げます。これらがまとまってユニークなフィンガープリントを構成します。

引用スタイル