<ul><li>Researchers propose using low-rank tensors to solve finite-horizon Markov Decision Processes (MDPs).</li><li>Policies and value functions in finite-horizon MDPs are not stationary, which poses challenges in high-dimensional MDPs.</li><li>The low-rank tensor representation enables scalable learning of optimal policies in finite-horizon MDPs.</li><li>Block-coordinate descent and block-coordinate gradient descent algorithms are introduced for solving the Bellman equations.</li></ul>

Solving Finite-Horizon MDPs via Low-Rank Tensors

Discover more