CEN : Cooperatively Evolving Networks
share
A finitely repeated game is a dynamic game in which a simultaneous game is played finitely many times. GANs contain two competing modules: the generator module is trained to generate new examples, and the discriminator module is trained to discriminate real examples from generated examples. Training procedure of GAN is a finitely repeated game in which each module tries to optimize it's error at every instance of simultaneous game in a non-cooperative manner. We observed that we can achieve more accurate training, if at each instance of simultaneous game the stronger module cooperate with weaker module and only weaker module only optimize it's error.
READ FULL TEXT