Quantum Memory in Support Vector Machines: An Empirical Study

Quantum Support Vector Machines (QSVMs) use quantum kernels to improve classification by mapping classical data into high-dimensional quantum feature spaces. This work explores the role of quantum memory in QSVMs by introducing a modified feature map that reuses quantum states during training. The proposed quantum memory kernel applies an additional layer of parameterized RY rotations, maintaining correlations between previously encoded states. Numerical simulations compare the performance of classical SVM, standard QSVM with the ZZFeatureMap, and QSVM with quantum memory across datasets of varying sizes. The results suggest that quantum memory helps reduce accuracy degradation as data complexity increases, indicating its potential for quantum machine learning.