Quantum recurrent neural networks for sequential learning.

Area of Science:

• Quantum Computing
• Artificial Intelligence
• Machine Learning

Background:

• The development of canonical quantum recurrent neural network (QRNN) models is crucial for advancing quantum deep learning.
• Noisy intermediate-scale quantum (NISQ) devices offer potential advantages for specific computational tasks.
• Existing QRNN models face limitations in terms of hardware requirements and accessibility on NISQ devices.

Purpose of the Study:

• To propose a novel, hardware-efficient quantum recurrent neural network (QRNN) model.
• To establish a canonical QRNN model suitable for near-term quantum devices.
• To demonstrate the practical applicability and superior performance of the proposed QRNN.

Main Methods:

• Construction of hardware-efficient quantum recurrent blocks (QRBs).
• Staggered stacking of QRBs to reduce quantum device coherence time requirements.
• Validation using diverse classical sequential datasets: meteorological indicators, stock prices, and text categorization.

Main Results:

• The proposed QRNN model shows significantly improved prediction and classification accuracy compared to classical recurrent neural networks (RNNs) and other quantum neural network (QNN) models.
• The QRNN effectively captures and predicts intricate details within temporal sequence data.
• The model's practical circuit structure and reduced coherence time requirements make it highly accessible for NISQ devices.

Conclusions:

• The developed QRNN is a strong candidate for a canonical QRNN model.
• The QRNN model offers a promising avenue for achieving quantum advantage in near-term applications.
• This research facilitates the practical implementation of quantum deep learning on current quantum hardware.