Data Rotation and Its Influence on Quantum Encoding.

Parametric quantum machine learning (QML) has been vastly studied over the last several years. These algorithms rely on hybrid implementations, where quantum methods define the models, and the parameters are update on classical devices. The encoding of classical data into quantum states within the Hilbert space is fundamental to training these hybrid models; this can be achieved in a number of ways. In this work, we focus on two of these methods, amplitude encoding and encoding via a second-order Pauli feature map. We compared their performances across two near-term QML models, quantum support vector classifier and variational quantum classifier. We found that amplitude encoding is significantly resilient to classical transformations of data. This work additionally introduces the concept of a rotation, applied to classical data as a preprocessing step. In our results, we observe that other encoding methods can significantly benefit from certain Cartesian rotations of the data. We expand this rotation to a larger n-D dataset and show the method’s performance.