Quantum Kernel-Based Long Short-Term Memory

This paper introduces QK-LSTM, a Quantum Kernel-Based Long Short-Term Memory architecture that fuses quantum computing with classical recurrent models for efficient sequence learning. Notable features include:

• Embedding input data into a quantum feature space, enabling compact representation of complex temporal dependencies.
• Reduced parameter count compared to classical LSTMs while maintaining equivalent modeling accuracy.
• Enhanced convergence and robustness, making the architecture ideal for edge devices and NISQ-limited quantum hardware.

Benchmark results show that QK-LSTM delivers competitive performance with greater efficiency, offering a practical path forward for quantum-enhanced sequence modeling in applications like natural language processing and real-time signal analysis.

Recommended citation: Hsu, Yu-Chao; Li, Tai-Yu; & Chen, Kuan-Cheng. (2025). "Quantum Kernel-Based Long Short-Term Memory." Proceedings of ICASSP 2025 – IEEE International Conference on Acoustics, Speech, and Signal Processing.
Download Paper