Comitê de redes neurais quânticas

Abstract

Deep Neural Networks open several possibilities to solve hard problems. The advancement in the use of quantum computation has allowed us to use quantum features which have no classical counterpart. It has brought forth many algorithms and techniques in the field of quantum machine learning (QML). One of the proposals was a binary quantum neural network (QBNN) which used an amplification method based on the well-known Grover’s algorithm search. Besides the fact of being a purely quantum method, it used an architecture selection mechanism as a meaningful approach. Despite its enhancements, one of the main disadvantages is the consumption of a lot of quantum computational resources. Therefore, We present a series of improvements from loading the data using a superposition instead of an original base encoding. Further, we also change the training process, replacing the costly Grover’s search with gradient descent optimization. It reduces the quantum computational loss, shrinking the number of operations and qubits. Moreover, we applied the concept of ensemble classification, instead of using a single specific quantum binary weight. Finally we show that it is possible to achieve a general model better using the Grover algorithm.