This research demonstrates improved classification accuracy in MNIST images using quantum convolutional neural networks, highlighting potential advantages over classical approaches.
Key Points
A quantum convolutional neural network implementation achieved 96.08% classification accuracy on MNIST images.
The encoding scheme reduced input dimensionality, eliminating the need for classical pre-processing techniques.
We introduced an automated framework to optimize quantum circuits based on expressibility and complexity features.
Results were validated on IBM's Heron r2 quantum processor, demonstrating clear advantages over traditional methods.