Efficient Risk-Averse Reinforcement Learning

Neural Information Processing Systems 

In risk-averse reinforcement learning (RL), the goal is to optimize some risk measure of the returns.