Quanvolutional Neural Networks

Quanvolutional Neural Networks

This PennyLane demo implements a Quanvolutional Neural Network (QuanvNN), a hybrid model that inserts a quantum convolution (quanvolution) layer into an otherwise classical convolutional neural network for image classification. Inspired by classical convolution, the quanvolutional layer slides a small quantum circuit over local patches of an input image: each patch is encoded into the circuit, the circuit (a fixed or random quantum transformation) processes it, and the measurement results produce several output feature channels, much like a classical convolutional filter but with a quantum kernel. The tutorial applies a quanvolutional layer to the MNIST handwritten-digit dataset, pre-processing the images through the quantum layer once to produce quantum-transformed feature maps, and then training a small classical network on those features. It uses PennyLane to define and run the per-patch quantum circuit and a classical framework for the surrounding network, and compares the quanvolutional model against a purely classical baseline. By embedding a quantum operation inside a familiar CNN architecture, the demo offers an accessible, image-focused example of hybrid quantum machine learning in PennyLane.

Run it

pip install -r requirements.txt
python demo.py

Source and license

Imported from demonstrations_v2/tutorial_quanvolution/demo.py in PennyLaneAI/demos at c52c0abeb5122218aa96b38eea848864cce7323f, under the Apache License 2.0. Original authors: Xanadu and the PennyLane community. The upstream LICENSE is included alongside this example.