[Done] Quantum-Image-Processing-from-visualization-to-classification_Qualition

[danielbultrini]

Project Name: Quantum Image Processing - embeddings - from visualization to classification Team Name: Qualition

Which challenges would you like to submit your project for?

• Quantum Computing today! (QPIXL is an algorithm that was published in May 2022 which we have implemented for qiskit and pennylane)
• Visualization Challenge (By performing various unitary transformations and entanglement between images we hope that it will give an intuitive 'view' into these operations)
• Hybrid Quantum-Classical Challenge (The project involves using both classical autoencoders as well as quantum ML to extend the capabilities of current machines)
• NVIDIA Challenge (By dealing with QML and images, we hope that the project will be of interest to NVIDIA)
• AWS Challenge (We use the embeddings to classify healthy cells from cancerous cells)

Project Link: https://github.com/Qualition/Quantum-Image-Processing/tree/e0624c17bba1d50ed5d292fb654985e2cc425158

Project description: The README goes through several embedding schemes we have come up with that are either simulator-friendly encodings or NISQ-friendly and implements them in both artistic and scientific settings. The main feature is that we have implemented an embedding for the recent FRQI-QPIXL framework (Amankwah et al., May 2022, https://www.nature.com/articles/s41598-022-11024-y ). we expanded it to include interactive demos, and examples including the use of a hybrid quantum-classical network for classifying a cancer dataset.

Then we have also developed a method for 'chunked' embedding where the image is split up and recombined into a compressed statevector. This method, which we call Distributed Amplitude Encoding is much easier on classical compute resources and allows for images as large as 4K to be processed with quantum operations being applied.

Finally, we have also looked at how well standard image embeddings perform in QML using a full quantum workflow - that is, direct quantum embedding, quantum autoencoder, and QNN classifier, and these can be seen in Autoencoder-QCNN.ipynb. Here the fashion-MINST data is used to benchmark performance. In the future, we hope to compare different embedding schemes and the performance of the same QNN with varying embeddings of the same dataset, which should be very interesting.