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)


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.

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
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


名前Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
9886 LNCS


Other25th International Conference on Artificial Neural Networks, ICANN 2016

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)


「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。