search

desireevl / quantum-game-theory

Playable quantum game theory games
https://quantum-game.desireevl.com/
Apache License 2.0
67 stars 13 forks source link

QIP Abstract Submission #7

Closed rajkk1 closed 4 years ago

rajkk1 commented 5 years ago

The Quantum Information Processing (QIP) Conference abstracts are due 13 Oct http://www.szpclab.com/qip2020#/submissions

JansenZZ commented 5 years ago

Tentative draft for title and abstract for QIP poster submission:

Title: Game-theoretic perspectives for multi-generator quantum GANs

Abstract: Generative adversarial networks (GANs) are powerful models in classical machine learning .... Its conventional architecture consists of a discriminator and a generator network (which endeavours to learn the latent distribution of some training dataset??).... the training of such a generative network is known to be equivalent to finding the Nash-equilibrium of a min-max game. Recent novel results have extended the GAN model to the quantum domain..... in this work, we examine the promise of quantum GANs from a game-theoretic perspective....(we are not sure what we will find, so I don't know what to say here exactly) We present experimental results on implementing quantum games on contemporary quantum devices..... (to be continued)

rajkk1 commented 5 years ago

To get to an optimal GAN, one has to find the Nash equilibrium of the minimax game between the Discriminator and Generator.

2 main concerns from my side:

  1. The game theory creator that we have designed creates n x n games, which does not include minimax games.
  2. The original QuGAN paper already proposes a quantum gradient descent method to try and obtain the minimax.

So here are 3 options that we could do:

  1. Go down the QuGAN avenue (e.g. look into multi-agent QuGANs as mentioned by Jansen), which would lead to moving our main focus away from the "standard" quantum game theory (and e.g. focus on minimax games).
  2. Continue to focus on quantum game theory framework we have been focusing on (including applications for QuGANs and other things) but probably not submit an abstract to QIP? (given that we don't have a clear idea of use cases/what we will add). We could always submit an abstract to the next interesting conference when we have a clearer idea :)
  3. Any other suggestions?

Let me know of your thoughts! Thanks team!