Learning the noise fingerprint of quantum devices

This paper shows that individual machines can be fingerprinted based on their aggregated results.

• It would be cool to expand these results:

1. Create a “Quantum Computer detector” that given a set of results, it returns the most likely computer it ran on. Ideally this algorithm would be multiclass, rather than using binary classification. We would like to add new quantum computers to the existing model without having to retrain the entire algorithm. This requires knowing how to add new classes to a multiclass algorithm
2. Use the quantum algorithm used as an input, rather than fixing the algorithm used.
   1. This requires mapping algorithms into something understandable by a Artificial Neural Network.
      1. Algorithms can be represented neatly as a DAG
      2. There ought to be solutions that map DAGs into a Artificial Neural Network input. This needs to be researched
   2. The training dataset should not use the same algorithm too many times: We want a general way to fingerprint for a generic algorithm.
3. Error correction: Can we add a per quantum computer map from the obtained results to the simulated results, in a way using Artificial Neural Networks to remove as much of the noise as possible?
   1. Can this be done as a many-shot correction ( i.e. mapping the full probability distribution ) ?
   2. Can this be done as a one-shot correction ( i.e. mapping an individual evaluation of the algorithm ) ?