Compact data encoding for data re-uploading quantum classifier

In the realm of quantum machine learning, different genres of quantum classifiers have been designed to classify classical data. Recently, a quantum classifier that features re-uploading the sample to be classified many times along the quantum circuit was proposed. Data re-uploading allows circumventing the limitations established by the no-cloning theorem. This quantum classifier has great potential in NISQ-era, because it requires very few qubits due to the special data encoding scheme it used. Previous work showed that even a single-qubit could constitute effective classifiers for problems with up to 4 dimensions. In this work, we focus our attention on the data encoding scheme of this quantum classifier, we propose an alternative way to encode the input sample in order to reduce by half the number of learnable parameters of the quantum circuit and simplify the computation, so the training time can be greatly shortened. Numerical results show that the new data encoding method achieves higher accuracy for high-dimensional data while using less parameters.