Abstract: This paper establishes a novel role for Gaussianmixture models (GMMs) as functional approximators of Qfunction losses in reinforcement learning (RL). Unlike the existing RL literature, where ...