We theoretically establish that our algorithms almost surely converge to locally optimal policies of our safe optimization framework.

Planning in long-horizon tasks is challenging in reinforcement learning (RL) (Co-Reyes et al., 2018;

Planning in long-horizon tasks is challenging in reinforcement learning (RL) (Co-Reyes et al., 2018;

For nonlinear systems, the problem is considerably harder.

A significant challenge in extending this success to mass production is the variation between instances of the problem, e.g.,

Existing work for this task has modeled the effect of multiple treatments independently, while estimating the joint effect has received little attention but comes with non-trivial challenges. In this paper, we propose a novel method for reliable off-policy learning for dosage combinations.