Sequential Counterfactual Decision-Making Under Confounded Reward

We investigate the limitations of random trials when the cause of interest is confounded with the effect by formalizing a counterfactual policy-space where the agent's natural predilection is input to a soft-intervention.