Fully Bayesian Recurrent Neural Networks for Safe Reinforcement Learning | Read Paper on Bytez