Improving Quantum Recurrent Neural Networks with Amplitude Encoding

Abstract:Recurrent Neural Networks (RNNs) are widely used to model complex temporal dependencies in sequential data. Quantum Recurrent Neural Networks (QRNNs) extend this idea to quantum machine learning by using quantum circuits inspired by classical recurrent architectures. In this paper, we propose three modifications to existing QRNN implementations that improve performance, generalization, and circuit efficiency. First, we introduce a simple preprocessing technique that augments amplitude encoded inputs with their prenormalized magnitudes, improving generalization on two financial data sets. Second, we show that a trained model can achieve comparable results with an approximate feature map rather than an expensive exact state preparation circuit. Third, we present a novel QRNN circuit architecture that is mathematically equivalent to the original model on ideal processors, but significantly reduces circuit depth. We demonstrate how these techniques can be put together to improve the performance of a QRNN when making predictions on real world financial data sets.

arXiv:2508.16784 [quant-ph]

(or arXiv:2508.16784v4 [quant-ph] for this version)

https://doi.org/10.48550/arXiv.2508.16784 Focus to learn more arXiv-issued DOI via DataCite

arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.

Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.

Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.