Global Optimality Guarantees for Policy Gradient Methods

The authors thank the anonymous referees for feedback that helped improve some aspects of the paper. A part of this work was completed when J. Bhandari was a research fellow at the Theory of Reinforcement learning program at the Simons Institute for the Theory of Computing, University of California, Berkeley. J. Bhandari thanks Peter Bartlett and Simons Institute for that opportunity and Gaurd Iyengar for love and support during the PhD program at Columbia University.