The GAME Algorithm proceeds from GMDH Multilayered Iterative Algorithm (MIA).

It evolves en ensemble of feedforward layered networks consisting of heterogeneous units (polynomial, sigmoid, gaussian, sin, perceptron, etc.). Natural evolution is used to optimize the topology of the network and to decide which combination of units is the most appropriate for given data. Several optimization methods (Quasi-Newton, Conjugate Gradient, Particle Swarm, Ant Colony, Differential Evolution, etc.) are competing to evolve most fit units by adjusting coefficients of their transfer function.

For more information please refer to  FAKE GAME thesis.