Quantum-Inspired Data Processing for Enhanced Machine Learning Performance.

Research codebase using MNIST that demonstrates how classical machine learning models can compete with quantum models through a novel hybrid quantum-classical feature engineering approach

1. Data Preparation

   • Load and preprocess the MNIST dataset.
   • Filter the dataset to include only the digits 0 and 9.
   • Reduce image dimensions using Principal Component Analysis (PCA).

2. Quantum Feature Encoding

   • Use quantum circuits to re-label the dataset.
   • Compute Projected Quantum Kernel (PQK) features via parameterized quantum circuits.
   • Calculate kernel matrices and perform eigenvalue decomposition to generate new labels that maximize the geometric separation between quantum and classical kernels.

3. Model Training and Comparison

   • Train a quantum-enhanced classical neural network on the PQK feature dataset.
   • Train a classical neural network on the original PCA-reduced dataset.
   • Compare the performance of both models (achieving > 90% accuracy on training data).

Installation

To install the required packages, run:

pip install tensorflow==2.3.1 tensorflow-quantum cirq sympy numpy matplotlib