Data Scanning Methods for Quantum-Classical Interface in Quanvolutional Neural Networks

Quantum machine learning has recently drawn a lot of attention of the research community. Particularly, hybrid quantum-classical architectures have been introduced to take advantage of advances in both classical computing and quantum computing. However, most existing quantum circuits are of the 1D type. For 2D contents, such as images, to be represented in quantum states, they must be converted to an 1D array. In this paper, we introduce some data scanning methods for 2D-to-D1 mapping. These methods are investigated in the context of quanvolutional neural networks. Through experiments, it is found that the proposed scanning methods provide significant gains compared to the default raster scan.