Speaker
Description
One key area where machine learning has shown particular promise is through the use of Convolutional Neural Networks (CNNs) in medical image analysis and recognition. In light of this, the advancements of quantum machine learning have introduced a variety of quantum ansatze for classical CNNs, called quantum convolutional neural networks (QCNNs), some of which are capable of capturing global correlations in input image data by leveraging entanglement. This work leverages HierarQcal, an open-source Python package that simplifies the process of quantum circuit design, composition, generation, scaling, and trainable parameter management. We use it to conduct a neural architecture search over different QCNN architectures that are trained on datasets from the MedMNIST benchmark collection. This package includes a range of circuit structure hyper-parameters that govern the connectivity of the parameterised two-qubit unitary gates that are used in the convolution and pooling layers of the QCNN architecture. This work investigates how varying the circuit structure hyper-parameters affects QCNN performance, exploring whether performance is determined not only by the choice of unitary gate and its number of parameters, but also by the connectivity of the gates throughout the circuit structure. Additionally, it aims to examine the specific QCNN models produced by each neural architecture search. This is expected to contribute to a more systematic understanding of QCNN design principles and demonstrate that careful tuning of both quantum operations and circuit connectivity is crucial for optimising learning in near-term quantum devices.
| Topic | Quantum algorithms for machine learning applications |
|---|---|
| Additional Topic | Quantum variational circuits and their applications |