| （英） |
In this presentation, we introduce our investigations about the relation between learning performance of a quantum machine learning model utilizing Hamiltonian dynamics and interaction networks of qubits. Especially, we pay attention to the periodicity of unitary maps in time evolution of the system. The periodicity yields a periodic feature in learning performance, which prevents the performance from improving. However, an interaction network, which is easily implementable, does not exhibit such a periodicity, and the performance grows up to one given by a random feature map. At last, we show that the non-existence of the periodicity of unitary maps can be guaranteed directly by the interaction network alone. |