Solving Finite-Horizon Discounted Non-Stationary MDPS

Abstract Research background Markov Decision Processes ( MDPs ) are a powerful framework for modeling many real-world problems with finite-horizons that maximize the reward given sequence of actions. Although such as investment and financial market where value decreases exponentially time, require introduction interest rates. Purpose This study investigates non-stationary finite-horizon discount factor to account fluctuations in rewards over time. methodology To consider authors define new nonstationary factor. First, existence an optimal policy proposed discounted is proven . Next, Discounted Backward Induction DBI algorithm presented find it. enhance their proposal, model used example MDP adaptive solve Results The method calculates values its expected total return consideration time money. Novelty No existing studies have before examined dynamic temporal rewards.