Moreover, since the SPO loss is not continuous nor convex in general [Elmachtoub and Grigas, 2021], which makesthe training ofaprediction model computationally intractable, Elmachtoub and Grigas [2021] introduced a novel convex surrogate loss, referred to as the SPO+ loss.

Predicting the most likely route from asource location to adestination is acore functionality inmapping services. Although theproblem hasbeenstudied inthe literature, two key limitations remain to be addressed.

We study minimax optimal reinforcement learning in episodic factored Markov decision processes (FMDPs), which are MDPs with conditionally independent transition components.

Building upon this "motion vocabulary", we perform language modeling on both motion and text in a unified manner, treating human motion as a

WW hinge loss and so on.