Curiosity, unobserved rewards and neural networks: On recent progress in building solid foundations for RL