I’m trying to implement DeepNash https://arxiv.org/abs/2206.15378.
Essentially, it is a way to get a policy model to converge to nash equilibrium in Imperfect Information Games without search, using only regularization. It is clear to me that this has myriad applications.
I am fairly certain I understand the paper and its prerequisites enough. I have already written a toy matrix-tree-game environment that is vectorized on the GPU, allowing for very fast simulation. I am also implementing a large grid based game that is visually intuitive and quite large in its number of information sets.
Would anybody like to hop on board?