Quantum Vision Transformer - Paper Implementation

[neogyk]

The Abstract:

The Vision Transformer is a very popular ML architecture in Computer Vision. It allows to perform the classification, regression and data generation in effective and efficient way. In this task we will implement the Quantum version of the vision transformer and will train it using the HEP data in order to simulate the EM shower in the detector.

Phases:

1. Review the existing application of the quantum vision transformer.
2. Implement the quantum vision transformer and training schema.
3. Use the HEP data for the quantum vision transformer to perform the classification task.
4. Use the HEP data to perform the generation.

Resources:

1. Quantum Vision Transformer
2. Quantum Vision Transformer
3. Quantum Vision Transformers for Quark-Gluon Classification
4. Hybrid Quantum Vision Transformers for Event Classification in High Energy Physics
   
5. End-to-End Quantum Vision Transformer:
   Towards Practical Quantum Speedup in Large-Scale Models

[TomerGoldfriend]

Sounds interesting @neogyk ! could you please provide more technical details on the implementation? will you use Classiq integration with PyTorch? What kind of quantum operations are you going to use (you can attach a layout of the quantum circuit for example, as well as the full hybrid network scheme).

[TomerGoldfriend]

@neogyk we are assigning you to this issue. Please note that we accept high-quality implementations to our repository and will be glad to accept a contribution that meets our standards.

Feel free to reach out to the community for any questions!
Good luck!

[TaliCohn]

Hi @neogyk, what is the status of this? Are you still working on the implementation?

[neogyk]

Dear @TaliCohn, I am still working, the code is not yet finalised.

[TaliCohn]

@neogyk, No problem, how long do you think you will need? I'll update the deadline

[neogyk]

@TaliCohn, tomorrow it's will be ready

[TaliCohn]

@neogyk any update here?

[neogyk]

Dear @TaliCohn . I have finished the implementation of the QVT, trained it using MNIST dataset. The backpropagation is very slow, it's very hard to achieve the results with the current efficiency.

[TaliCohn]

@TomerGoldfriend Please review the PR