Multi-class classification using quantum transfer learning

Image classification is one of the most important machine learning tasks, especially in this digital era. Though there exists classical algorithms which have performed quite well in multi-class classification tasks, classification using quantum architectures have mostly been limited to 2 or 3 classes. As the number of classes increased, the existing architectures did not achieve good accuracy. In this work, we aim to classify the MNIST dataset into 10 corresponding classes, using classical-to-quantum transfer learning. We performed both binary as well as multi-class classification using the hybrid architecture which yielded a maximum accuracy of approximately 100 and 90.4% respectively.