Near Optimal Policy Optimization via REPS

Authors: Aldo Pacchiano, Jonathan N Lee, Peter Bartlett, Ofir Nachum

NeurIPS 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Theoretical In this paper we aim to fill this gap by providing guarantees and convergence rates for the sub-optimality of a policy learned using first-order optimization methods applied to the REPS objective. We first consider the setting in which we are given access to exact gradients and demonstrate how near-optimality of the objective translates to near-optimality of the policy. We then consider the setting of stochastic gradients and introduce a technique that uses generative access to the underlying Markov decision process to compute parameter updates that maintain favorable convergence to the optimal regularized policy.
Researcher Affiliation Collaboration Aldo Pacchiano Microsoft Research apacchiano@microsoft.com Jonathan Lee Stanford University jnl@stanford.edu Peter L. Bartlett UC Berkeley peter@berkeley.edu Google ofirnachum@google.com
Pseudocode Yes Algorithm 1 Relative Entropy Policy Search [Sketch]. Input: Initial iterate v0, accuracy level > 0, gradient optimization algorithm O. ... Algorithm 2 Biased Gradient Estimator Input Number of samples t. ... Algorithm 3 Biased Stochastic Gradient Descent Input Desired accuracy , learning rates { t}1 t=1, and number-of-samples function n : N ! N .
Open Source Code No Did you include the code, data, and instructions needed to reproduce the main experi- mental results (either in the supplemental material or as a URL)? [N/A]
Open Datasets No The paper is theoretical and does not mention using any specific datasets for training or provide access information for any dataset.
Dataset Splits No The paper is theoretical and does not mention specific dataset splits for training, validation, or testing.
Hardware Specification No Did you include the total amount of compute and the type of resources used (e.g., type of GPUs, internal cluster, or cloud provider)? [N/A]
Software Dependencies No The paper does not provide specific software dependencies with version numbers.
Experiment Setup No The paper is theoretical and does not describe a concrete experimental setup with hyperparameters or system-level training settings.