Quantum Transfer Learning for Lymph Node Metastases Detection

The Quantum GoogLeNet model. The quantum layer: the QAOA-inspired ansatz embedding, the particle-conserving entangler, and the expectation value of the Pauli Z operator.

Transfer learning may make training on a particularly distinguishable dataset easier. It enables several elements of a pre-trained model to be used as the foundation of a new model's architecture. More importantly, we can adopt this approach in quantum machine learning as well. In this project, we seek to implement quantum transfer learning using an ImageNet-pre-trained model, which will be used on the PCam dataset to tackle the lymph node metastases detection problem. The pre-trained model is GoogLeNet (i.e., Inception V1), and the classifier uses hybrid classical-quantum fully connected layers. Typically, quantum layers are made up of embedding, quantum circuits, and measurement. The embedding and quantum circuits are built upon the QAOA-inspired ansatz and particle-conserving entangler, respectively.

Experiment

Consider exploring this notebook to conduct the experiment by yourself.

Result

Quantitative Result

The quantitative results are outlined in the following table.

Test Metric	Score
Accuracy	80.29%
Loss	0.464

Accuracy and Loss Curves

The model's loss curve on the train and validation sets.

The model's accuracy curve on the train and validation sets.

Qualitative Result

This 3×3 image grid presents the qualitative result.

.

Citation

If you find this repository useful for your research, please cite it:

@misc{quantum-transfer-learning-metastases,
   title = {Quantum Transfer Learning for Lymph Node Metastases Detection},
   url = {https://github.com/reshalfahsi/quantum-transfer-learning-metastases},
   author = {Resha Dwika Hefni Al-Fahsi},
}

Credit

• Going deeper with convolutions
• PatchCamelyon (PCam)
• Rotation Equivariant CNNs for Digital Pathology
• Transfer learning in hybrid classical-quantum neural networks
• Quantum embeddings for machine learning
• Quantum algorithms for electronic structure calculations: particle/hole Hamiltonian and optimized wavefunction expansions
• PennyLane: Automatic differentiation of hybrid quantum-classical computations
• Turning quantum nodes into Torch Layers
• PyTorch Lightning