Open andreamari opened 2 years ago
@andreamari Myself and a few others are reading the paper and learning about this. So we can take this issue.
@isaac-gs, great!
Quick update, we are almost done with a refactor of the existing CDR code that should help make it easier to add new CDR approaches in the future.
We also have an MCMC implementation that will be added in a separate PR, but there's still some additional work to do there as well.
Thanks @isaac-gs! There is no hurry, take your time. As you already mentioned, if possible, please try to split it into small intermediate PRs. Feel free to join our weekly community calls on Discord (http://discord.unitary.fund/) if you want to discuss about this synchronously.
The way in which training circuits are generated for the CDR technique is mentioned in the original paper in which CDR was introduced Czarnik et al. arXiv:2005.10189 but was not used in the associated numerical experiments. In the same paper a different method was proposed in which training circuits are generated with a Markov Chain Monte Carlo approach. This is the method that was actually used in all the numerical simulations of the paper.
It would be great to have the Markov Chain Monte Carlo method implemented in Mitiq as an additional way of generating training circuits.