Fully Quanvolutional Networks for Time Series Classification
Keywords: quanvolutional neural networks, quanvolution, time series classification
Abstract: Quanvolutional neural networks have shown promise in areas such as computer vision and time series analysis. However, their applicability to multi-dimensional and diverse data types remains underexplored. Existing quanvolutional networks heavily rely on classical layers, with minimal quantum involvement, due to inherent limitations in current quanvolution algorithms. In this study, we introduce a new quanvolution algorithm that addresses previous shortcomings related to performance, scalability, and data encoding inefficiencies. Specifically targeting time series data, we propose the Quanv1D layer, which is trainable, capable of handling variable kernel sizes, and can generate a customizable number of feature maps. Unlike previous implementations, Quanv1D can seamlessly integrate at any position within a neural network, effectively processing time series of arbitrary dimensions. Our chosen ansatz and the overall design of Quanv1D contribute to its significant parameter efficiency and inherent regularization properties. In addition to this new layer, we present a new architecture called Fully Quanvolutional Networks (FQN), composed entirely of Quanv1D layers. We tested this lightweight model on 20 UEA and UCR time series classification datasets and compared it against both quantum and classical models, including the current state-of-the-art, ModernTCN. On most datasets, FQN achieved accuracy comparable to the baseline models and even outperformed them on some, all while using a fraction of the parameters.
Primary Area: learning on time series and dynamical systems
Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.
Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.
Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.
No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.
Submission Number: 7311
Loading