qibo opt optimizers and qibo ml optimizers

[shangtai]

Hi Everyone,

We are also developing a Qibo Opt repo. The main target audience is meant for people from classical optimization background to use Qibo easily. To do so, we will first prepare some basic standard classical Optimization Classes (such as QUBO) that can be used for quantum computing and also work out some applications classes.

Furthermore, I am wondering if we can include some Optimizers as well. I don't see why we should included those scipy optimizers as we can already directly call those functions easily.

1. I wonder if there are some optimizers that would be helpful for Qibo ML since one of the main applications of optimizers would be for the purpose of parameters tuning in machine learning. Do let me know if there are any suggestions to be implemented that would find applications in machine learning.

2. The applications of optimizers is beyond machine learning, it is used in operations research and finance. I think it's alright to have optimizers in Qibo ML to make thing convenient as well but it would be great to have optimizers in Qibo Opt as well. Do let me know if you have any suggestions.

@MatteoRobbiati Do let me know if you have any suggestions.

[MatteoRobbiati]

Hi @shangtai! I am sorry for the delay and thanks to tag me here, I didn't notice the issue!
What is the status of the Qibo-opt repo? I searched for it in the organization and I wasn't able to find it!

Anyway, I totally agree with you when saying optimizers are beyond QML. For now, we were experimenting in qiboml a different version of the optimizers we have in qibo.

Since we have plans in mind for the optimization structure, it could be a good idea to setup a meeting to coordinate all these efforts. In particular since other people are interested in contributing (e.g. the Quantum Technology Initiative at CERN).

As soon as I will be back from a school, I'll set up a meeting! :)

In the meanwhile, me, @Edoardo-Pedicillo and @alecandido are having a look to this qiboml repo in order to move it to qiboteam ASAP.

[shangtai]

Hi @MatteoRobbiati , thanks for the reply.

As of now, currently, Qibo Opt is still in the Nascent stage where I am still brainstorming what should be included and what is an appropriate structure.

• Some basic components that I think is required:

  I think I need a QUBO class since that's the language that classical optimization folks use. I am also including a linear class for now so that a method known as Augmented Langrangian approach can be used easily. I also intend to implement a variant of ADMM algorithm where hybrid quantum computing and classical computing can be performed. After which, I shall include some applications. Over time, I will include more applications and algorithms.

• Thoughts on optimizers

  Optimizers: Currently I am thinking how should we classify the optimizers.

  • Some optimizers are commonly used, especially those that are in scipy packages, I believe this should be in the main repo and doesn't require us to include them in Qibo Opt and Qibo ML.

  • Some optimizers might be only used in ML, for those I believe they should be in Qibo ML.

  • Some optimizers might have wider applications but they are not commonly used. I believe those should be in Qibo Opt.

For example, I am wondering where should IMFIL be positioned that will benefit Qibo Users most.