Reproduce “Bridging Classical and Quantum Machine Learning: Knowledge Transfer From Classical to Quantum Neural Networks Using Knowledge Distillation” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Mohammad Junayed Hasan, M.R.C.Mahdy

Researchers at Johns Hopkins University and North South University developed a knowledge distillation framework to transfer learned representations from classical neural networks to quantum neural networks. This approach consistently improved the accuracy of quantum models on image classification tasks, yielding average gains of 3.74% on MNIST, 2.36% on FashionMNIST, and 2.58% on CIFAR10, while operating with significantly fewer parameters than classical teacher models.

Key insights

  • Knowledge can be effectively transferred from classical neural networks to quantum neural networks using distillation, leading to consistent performance improvements across various image classification benchmarks.
  • The choice of dimensionality reduction (fully connected layers) and data encoding (amplitude encoding) significantly impacts QNN performance, with these methods outperforming alternatives by substantial margins.
  • Quantum student models achieve competitive performance with significantly fewer parameters compared to their classical teacher counterparts, demonstrating parameter efficiency critical for NISQ-era quantum computing.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.