[19a-331-7~12] 10.3 Spin devices, magnetic memories and storages

▲ [19a-331-12] Non-linearity in reservoir computing with nanomagnet array

〇Taishi Furuta¹, Nomura Hikaru¹, Goto Minori^1,3, Miwa Shinji^1,2,3, Kuwabiraki Yuki¹, Nakatani Ryoichi¹, Suzuki Yoshishige^1,3 (1.Osaka Univ., 2.The Univ. of Tokyo, 3.CSRN-Osaka)

Keywords:nanomagnet, recurrent neural network, reservoir computing

Recurrent neural network (RNN) is a mathematical model for machine learning, which emulates the nerve system in human brain. RNN consists of many nodes which keep information as a state of the nodes, and the nodes interact with each other. Fig. 1 shows a concept of the RNN. Each node in the middle layer shows a non-linear response via the interaction between the nodes. The state of the nodes is updated recursively, i.e. the state is determined by the present input and the previous state. Therefore this property can be used for computing a time series data. Recently, to improve a performance of the RNN, reservoir computers (RCs) with physical system have been reported. The RC is one of the RNN and have simple architecture. In this research, we introduce a RC with nanomagnet array which calculate a state via magneto static interaction between the nanomagnets, and evaluate a non-linearity in the nanomagnet RC (NM-RC)