Wasserstein Entropic Value-at-Risk for Robust Agent Learning
August 20, 2026
The proposed Wasserstein entropic value-at-risk uses an optimal-transport ball to account for catastrophic risks that standard entropic measures miss. It provides a closed-form robust dynamic-programming operator where caution levels adjust based on belief entropy.
HOW THIS AFFECTS YOU
●
researcherYou can implement safer reinforcement learning agents that hedge against extreme tail risks using optimal transport.