From cce61a537243106d349b9ed3bb5c695ca8effd2c Mon Sep 17 00:00:00 2001 From: Kaiyang Guo Date: Tue, 11 Oct 2022 20:29:51 +0800 Subject: [PATCH] update README.md --- README.md | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) diff --git a/README.md b/README.md index c752a0ba..6e00256e 100644 --- a/README.md +++ b/README.md @@ -171,7 +171,7 @@ Safe exploration is a challenging and important problem in model-free reinforcem ## [Model-Based Offline Reinforcement Learning with Pessimism-Modulated Dynamics Belief](./PMDB) Code associdated to: [Model-Based Offline Reinforcement Learning with Pessimism-Modulated Dynamics Belief](https://nips.cc/Conferences/2022/Schedule?showEvent=54842) accepted -at **NeurIPS22** conference.. +at **NeurIPS22** conference. #### Abstract Model-based offline reinforcement learning (RL) aims to find highly rewarding policy, by leveraging a previously @@ -183,7 +183,7 @@ through reward penalty may incur unexpected tradeoff between model utilization a instead maintain a belief distribution over dynamics, and evaluate/optimize policy through biased sampling from the belief. The sampling procedure, biased towards pessimism, is derived based on an alternating Markov game formulation of offline RL. We formally show that the biased sampling naturally induces an updated dynamics belief with -policy-dependent reweighting factor, termed \emph{Pessimism-Modulated Dynamics Belief}. To improve policy, we devise an +policy-dependent reweighting factor, termed *Pessimism-Modulated Dynamics Belief*. To improve policy, we devise an iterative regularized policy optimization algorithm for the game, with guarantee of monotonous improvement under certain condition. To make practical, we further devise an offline RL algorithm to approximately find the solution. Empirical results show that the proposed approach achieves state-of-the-art performance on a wide range of benchmark tasks.