Abstract
The integration of quantum computing into classical machine learningarchitectures has emerged as a promising approach to enhance model efficiencyand computational capacity. In this work, we introduce the Quantum Kernel-BasedLong Short-Term Memory (QK-LSTM) network, which utilizes quantum kernelfunctions within the classical LSTM framework to capture complex, non-linearpatterns in sequential data. By embedding input data into a high-dimensionalquantum feature space, the QK-LSTM model reduces the reliance on largeparameter sets, achieving effective compression while maintaining accuracy insequence modeling tasks. This quantum-enhanced architecture demonstratesefficient convergence, robust loss minimization, and model compactness, makingit suitable for deployment in edge computing environments and resource-limitedquantum devices (especially in the NISQ era). Benchmark comparisons reveal thatQK-LSTM achieves performance on par with classical LSTM models, yet with fewerparameters, underscoring its potential to advance quantum machine learningapplications in natural language processing and other domains requiringefficient temporal data processing.