tail-risk

The paper introduces quantile propagation planning (QPP), which propagates a fixed set of disturbance quantiles through dynamics to evaluate both mean and tail risk under a limited model-evaluation budget, showing reduced tail cost and constraint violations over mean-rollout planning on overload and inventory tasks.

This paper extends an Extreme Value Theory approach for CVaR optimization in reinforcement learning to high-dimensional neural-network policies, obtaining a differentiable tail-risk objective computable via automatic differentiation, and shows improved hedging performance under extreme scenarios compared to empirical-CVaR-trained policies.