Quantum Convolutional Model

This project integrates quantum computing with classical machine learning to build a hybrid Quantum Convolutional Neural Network (QCNN) for digit classification using the MNIST dataset. It leverages PennyLane for quantum circuit simulation and TensorFlow/Keras for training a classical neural network on quantum-processed features.

Features:

• Quantum Feature Extraction: A 2×2 sliding window is used to feed image patches into a 4-qubit quantum circuit, extracting 4 quantum features per patch (i.e., quantum convolution).

• Preprocessing Option: Quantum-transformed image data is optionally precomputed and stored to reduce training time on subsequent runs.

• Model Architecture: A simple feedforward neural network is trained on the 14×14×4 quantum feature maps.

• Prediction on Custom Images: Custom handwritten digits (in image format) can be passed through the same quantum pipeline and classified by the trained model.

• Visualization: Quantum channel outputs (Q0–Q3) are visualized alongside original input images.

Tech Stack:

• Quantum Layer: PennyLane, NumPy

• Classical Layer: TensorFlow/Keras

• Image Processing: PIL, Matplotlib

• Dataset: MNIST

Graphs

Quantum-processed output

Training Performance

Prediction visualisztion

Deployment

To deploy this project run

   git clone https://github.com/EmaadAkhter/Quantum-Convolutional-Model.git  

Go to the project directory

  cd Quantum-Convolutional-Model
  cd code

Run the setup file

  sh setup.sh

Run the python script

  python main.py

Acknowledgements

• Quanvolutional Neural Networks

• pennylane