A randomized transformation for data processing on quantum computers

Quantum state preparation circuits for loading classical data into quantum computers significantly influence the performance and complexity of variational quantum algorithms within hybrid quantum-classical systems. Since real-world data is often high-dimensional and mapping it to a quantum state is costly and non-trivial, it is transformed on classical computers to overcome limitations regarding quantum register size and circuit complexity. We present a data transformation method using Bloom filters to represent classical data for quantum state preparation circuits. Further, the paper illustrates that tiny fragments of this classical data encoding can be used for quantum machine learning to make an ensemble of the resulting quantum models surprisingly powerful. Due to the representation of the transformed data as a fragmented bit array, the quantum state preparation circuit only relies on a single rotational gate per qubit and small quantum registers. We demonstrate our approach and its representation power in a series of simulations. The simulations indicate that the randomized transformation provides diversity for ensemble models and that even small bit arrays with high error rates in data representation are sufficient for binary classification tasks.