Deep Learning Across Games (online resource)

Condorelli, D. and; Furlan, M. Deep Learning Across Games arXiv:2409.15197 (2024).


Input: a bimatrix game (A, B), where A is the payoff matrix for the row player and B for the column player.

Output: a strategy profile (x, y), where x is the mixed strategy selected by the row player network, and y by the column player network.

Example input:

A =1-1
-11
B =01
10

Matrices A and B must be either 2 x 2 or 3 x 3 and match the selected networks' dimensions. Payoff values can be signed integers or decimals. For zero-sum games, B is set to –A; for symmetric games, B is set to AT. Optionally, A and B can be normalized to have mean zero and norm n.

Evaluation on a list of known games can be found here.


Neural networks: