11
$\begingroup$

I already have a functioning $Q(\lambda)$ implementation for a single agent working on a dynamic pricing problem with the goal of maximizing revenue. The problem that I'm working with, however, involves several different products that are replacements for each other, so dynamically pricing them all with independent learners seems incorrect, because the price of one influences the reward of the other. The goal would be to dynamically price them all so as to maximize the sum of each individual revenue.

I've been doing some research to try to find something that applies reinforcement learning in this way, but many multi-agent implementations I have found focus more on competitive games than cooperative, or they assume incomplete knowledge of other agents (I would have complete knowledge of each agent in this scenario). Are there any well-researched/documented applications of cooperative learning in this way?

Improve this question
$\endgroup$

2 Answers 2

Reset to default
1
$\begingroup$

You can look these paper. First one is quite related to your task.
http://icamt2016.org/papers/SS-LTMLBDA-06-05.pdf http://researcher.watson.ibm.com/researcher/files/us-kephart/icml00_qrt.pdf

Improve this answer
$\endgroup$
1
  • 1
    $\begingroup$ Could you provide more information like the paper titles, author names, and dates so we can look them up in the event of dead links? The first link is dead, unless it is simmply does not work for me specifically. $\endgroup$
    – Walid Mujahid وليد مجاهد
    Commented Apr 11, 2021 at 3:29
0
$\begingroup$

All in all, what you are trying to reach is Pareto efficiency.

In order to make it cooperative, you need to define a single reward function that is shared by all the players (it could be a function that combines in some way individual reward functions).

Somehow, you need to weight the rewards that you obtain from one product with respect to the others.

Improve this answer
$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.