qnn_visualization

For QISKIT Hackathon Korea

Authors

• Joon Suk Huh
• Dongmin Kim
• Yisak Kim

Visualistic evaluations of quantum neural network layers

Original paper : "Transfer learning in hybrid classical-quantum neural networks" arXiv:1912.08278 (2019).

Motivations

• The number of literatures on Quantum Machine Learning (QML) is rapidly increasing recently [1].
• Especially, a method of QML called Quantum Neural Network (QNN) has been intensively investigated [2-4].
• However, advantages of QNN are yet to be fully understood both theoretically and empirically.

Objectives

• We would like to investigate the efficacy of QNNs by using various visual interpretation methods developed for classical Deep Neural Networks (DNN).
• Our model of choice is the Hybrid Quantum-Classical Neutral Network (Hybrid-QCNN) from [4].
• This model uses pre-trained DNN (ResNet18, trained over ImageNet) followed by ‘dressed’ 4-qubit QNN.
• The authors of [4] trained parameters of ‘dressed’ 4-qubit QNN part over CIFAR-10 dataset (which is different from ImageNet).

Contents

• classical_cat_dog.pt : classical neural network

• quantum_cat_dog.pt: neural network w/ quantum fc layer

• quantum_deep_dream.ipynb : quantum network deep dream [5]

• quantum_network_saliency_map.ipynb : quantum network saliency map [6]

• t-sne_layer_by_layer.ipynb : quantum network layer-wise t-sne [7]

• quantum_layer_activation_visualization.ipynb : quantum network layer activation manifold and whitening

• quantum_loss_landscape.ipynb : quantum network training loss landscapes [8]

[1] Biamonte, Jacob, et al. "Quantum machine learning." Nature 549.7671 (2017): 195-202.

[2] Jeswal, S. K., and S. Chakraverty. "Recent developments and applications in quantum neural network: a review." Archives of Computational Methods in Engineering 26.4 (2019): 793-807.

[3] Beer, Kerstin, et al. "Training deep quantum neural networks." Nature communications 11.1 (2020): 1-6.

[4] Mari, Andrea, et al. "Transfer learning in hybrid classical-quantum neural networks." Quantum 4 (2020): 340.

[5] Mordvintsev, Alexander; Olah, Christopher; Tyka, Mike (2015). "DeepDream - a code example for visualizing Neural Networks". Google Research. Archived from the original on 2015-07-08.

[6] Simonyan, Karen, Andrea Vedaldi, and Andrew Zisserman. "Deep inside convolutional networks: Visualising image classification models and saliency maps." arXiv preprint arXiv:1312.6034 (2013).

[7] Hinton, Geoffrey, and Sam T. Roweis. "Stochastic neighbor embedding." NIPS. Vol. 15. 2002.

[8] Li, Hao, et al. "Visualizing the loss landscape of neural nets." arXiv preprint arXiv:1712.09913 (2017).