REVIEW 3 cited by
Neural Policy Gradient Methods: Global Optimality and Rates of Convergence
Not yet reviewed by Pith; the record is open.
This paper has not been read by Pith yet. Machine review is queued; the pith claim, tier, and objections will appear here once it completes.
SPECIMEN: schema-true, not a live event
T0 review · schema-true
One-sentence machine reading of the paper's core claim.
pith:XXXXXXXX · record.json · timestamp
read the original abstract
Policy gradient methods with actor-critic schemes demonstrate tremendous empirical successes, especially when the actors and critics are parameterized by neural networks. However, it remains less clear whether such "neural" policy gradient methods converge to globally optimal policies and whether they even converge at all. We answer both the questions affirmatively in the overparameterized regime. In detail, we prove that neural natural policy gradient converges to a globally optimal policy at a sublinear rate. Also, we show that neural vanilla policy gradient converges sublinearly to a stationary point. Meanwhile, by relating the suboptimality of the stationary points to the representation power of neural actor and critic classes, we prove the global optimality of all stationary points under mild regularity conditions. Particularly, we show that a key to the global optimality and convergence is the "compatibility" between the actor and critic, which is ensured by sharing neural architectures and random initializations across the actor and critic. To the best of our knowledge, our analysis establishes the first global optimality and convergence guarantees for neural policy gradient methods.
Forward citations
Cited by 3 Pith papers
-
Geometry of Neural Reinforcement Learning in Continuous State and Action Spaces
For wide two-layer linearized neural policies in deterministic continuous RL, the locally attainable states concentrate on a manifold of dimension at most 2da+1, independent of the state dimension.
-
Adaptive Partitioning and Learning for Stochastic Control of Diffusion Processes
APL-Diffusion achieves a regret bound in K whose exponent is governed by a new zooming dimension tailored to unbounded diffusions, recovering Sinclair et al.'s rate as the initial-state moment p goes to infinity.
-
The Actor-Critic Update Order Matters for PPO in Federated Reinforcement Learning
Reversing PPO's update order in federated learning, actor before critic, removes the pairwise critic-difference term from the convergence bound and improves performance on heterogeneous control tasks.
Discussion (0). Continue with ORCID to comment.