The Quantum Convolutional Neural Network (QCNN) is a quantum circuit model inspired by the architecture of Convolutional Neural Networks (CNNs). CNNs are successful because they do not need manual feature design and can learn high-level features from raw data. Neural Architecture Search (NAS) builds on this success by learning network architecture and achieves state-of-the-art performance. However, NAS requires the design of a search space, which is currently not possible for QCNNs as no formal framework exists to capture its design elements. In this work, we provide such a framework by using techniques from NAS to create a hierarchical representation for QCNN architectures. Using this framework, we generate a family of popular QCNNs, those resembling reverse binary trees. We then evaluate this family of models on a music genre classification dataset, GTZAN, showing that alternating architecture has a greater impact on model performance than other modelling components, such as the choice of unitary ansatz and data encoding. Our framework provides a way to improve model performance without increasing complexity and to jump around the cost landscape to avoid barren plateaus. Finally, we implement the framework as an open-source Python package to enable dynamic QCNN creation and facilitate QCNN search space design for NAS.

翻译：量子卷积神经网络（QCNN）是一种受卷积神经网络（CNN）架构启发的量子电路模型。CNN之所以成功，在于其无需手动设计特征，即可从原始数据中学习高层特征。神经架构搜索（NAS）在此基础上通过学习网络架构进一步提升了性能，并实现了最先进的结果。然而，NAS需要设计搜索空间，但当前由于缺乏能够捕捉QCNN设计要素的形式化框架，无法对QCNN进行此类操作。本文利用NAS技术，提出了一套用于构建QCNN架构层级表示的形式化框架。基于该框架，我们生成了常见QCNN系列——即近似反向二叉树结构的模型。随后，在音乐流派分类数据集GTZAN上评估了该系列模型，结果表明：相较于其他建模组件（如酉ansatz选择与数据编码方式），交替架构对模型性能的影响更为显著。该框架提供了一种在不增加复杂度的情况下提升模型性能的方法，并能在代价地貌中跳跃以避免贫瘠高原。最后，我们将该框架实现为开源Python包，以支持动态QCNN创建，并促进面向NAS的QCNN搜索空间设计。