Minimum information divergence of Q-functions for dynamic treatment resumes

This paper aims at presenting a new application of information geometry to reinforcement learning focusing on dynamic treatment resumes. In standard framework learning, Q-function is defined as the conditional expectation reward given state and an action for single-stage situation. We introduce equivalence relation, called policy equivalence, in space all Q-functions. A class divergence every stage. The main objective propose estimator optimal function by method minimum based dataset trajectories. particular, we discuss $$\gamma $$ -power that shown have advantageous property such between policy-equivalent Q-functions vanishes. essentially works seek policy, which discussed semiparametric model Q-function. specific choices power index give interesting relationships value function, geometric harmonic means numerical experiment demonstrates performance context regimes.