Speaker
Description
Quantum convolutional neural networks (QCNNs) provide a framework for quantum representation learning through alternating convolution and pooling operations. Although many QCNN architectures remain efficiently classically simulable, they provide a framework for studying quantum representation learning on near-term quantum devices. While quantum autoencoders (QAEs) have been proposed for reconstruction tasks, the influence of latent-space architecture on reconstruction-based anomaly detection remains largely unexplored. We investigate QCNN-based QAE variants for semi-supervised anomaly detection using feature-extracted time-series exoplanet data.
The first variant, the persistent quantum autoencoder (pQAE), progressively concentrates information onto a reduced set of latent qubits while retaining all qubits throughout the QCNN hierarchy. The second, the bottleneck quantum autoencoder (bQAE), explicitly compresses information into a latent quantum state before reconstruction using a mirrored decoder operating on additional qubits. The corresponding circuits are shown in the Figure where (a) depicts the persistent architecture and (b) the bottleneck architecture. Multiple latent-space sizes are investigated to study the trade-off between compression, qubit resources, and reconstruction fidelity.
Both variants employ angle encoding together with trainable convolution, pooling, and mirrored decoding operations, with mean squared reconstruction error serving as both the optimization objective and anomaly score. We evaluate how latent-space design influences reconstruction-based anomaly detection by comparing latent-space sizes across the two architectures. The bottleneck quantum autoencoder consistently outperforms the persistent architecture, suggesting that explicit latent-space compression provides a more effective representation for this task. Although supervised classical and quantum benchmarks achieve stronger overall performance, demonstrating quantum advantage is not the objective of this work.
| Topic | Quantum variational circuits and their applications |
|---|---|
| Additional Topic | Encoding and processing of data in quantum systems |