Exploring the Capabilities of Quantum Support Vector Machines for Image Classification on the MNIST Benchmark

Quantum computing is a rapidly growing field of science with many potential applications. One such field is machine learning applied in many areas of science and industry. Machine learning approaches can be enhanced using quantum algorithms and work effectively, as demonstrated in this paper. We present our experimental attempts to explore Quantum Support Vector Machine (QSVM) capabilities and test their performance on the collected well-known images of handwritten digits for image classification called the MNIST benchmark. A variational quantum circuit was adopted to build the quantum kernel matrix and successfully applied to the classical SVM algorithm. The proposed model obtained relatively high accuracy, up to 99%, tested on noiseless quantum simulators. Finally, we performed computational experiments on real and recently setup IBM Quantum systems and achieved promising results of around 80% accuracy, demonstrating and discussing the QSVM applicability and possible future improvements.