Title:

OBR Newton’s Method a Faster Solver for Markov Games

Reinforcement Learning is a popular variant of machine learning that has cropped up as tasks given to AI have become more complex. It’s a framework that teaches a model to complete a task by trying it, and optimizing its decisions around the rewards it receives. In circumstances where there’s uncertainty about the dynamics of an environment it can be helpful to have an adversarial environment which seeks to make the uncertain dynamics of the environment as bad as possible to allow the model to find a safer solution with regard to the unknowns. A simple version of this concept is a Markov Game where the adversarial environment is just another agent whose rewards are the opposite of the original agent. When these two agents are at a stalemate the game is solved. A variety of methods for solving these games are explored and compared.

Name

Comment*

Comments are viewable only by submitter

Captcha*