GAN + Active Learning on top of Reasoning

Adversarial training of the verifier along with reasoner

Potentially in one model with learning to verify