Soft Actor-Critic: Off-Policy Maximum Entropy Deep Reinforcement Learning with a Stochastic Actor

2019-07-15 22:23:02

Paper: https://arxiv.org/pdf/1801.01290.pdf or Updated Version: https://arxiv.org/pdf/1812.05905.pdf

Project: https://sites.google.com/view/soft-actor-critic or https://sites.google.com/view/sac-and-applications/

TensorFlow: https://github.com/haarnoja/sac

PyTorch: https://github.com/vitchyr/rlkit

Demo video: https://www.youtube.com/channel/UCxXt8Br3-wyluz9Q08-fsaA

Good Related Blog: https://zhuanlan.zhihu.com/p/70360272

==== Video Related Tutorials (A2C, A3C): 

A brief review of Actor-Critic Algorithms:   https://www.youtube.com/watch?v=aODdNpihRwM

CS885 Lecture 7b: Actor Critic:        https://www.youtube.com/watch?v=5Ke-d1Itk3k

DRL Lecture 6: Actor-Critic:          https://www.youtube.com/watch?v=j82QLgfhFiY&t=27s

Build an A2C agent that learns to play Sonic with Tensorflow (tutorial):   https://www.youtube.com/watch?v=GCfUdkCL7FQ

Reinforcement Learning 6: Policy Gradients and Actor Critics (Deep Mind):    https://www.youtube.com/watch?v=bRfUxQs6xIM&t=27s

Actor Critic (A3C) Tutorial:         https://www.youtube.com/watch?v=O5BlozCJBSE

Actor Critic Algorithms:            https://www.youtube.com/watch?v=w_3mmm0P0j8&t=2s

 

==

Soft Actor-Critic: Off-Policy Maximum Entropy Deep Reinforcement Learning with a Stochastic Actor的更多相关文章

  1. 18 Issues in Current Deep Reinforcement Learning from ZhiHu

    深度强化学习的18个关键问题 from: https://zhuanlan.zhihu.com/p/32153603 85 人赞了该文章 深度强化学习的问题在哪里?未来怎么走?哪些方面可以突破? 这两 ...

  2. (zhuan) Deep Reinforcement Learning Papers

    Deep Reinforcement Learning Papers A list of recent papers regarding deep reinforcement learning. Th ...

  3. (转) Deep Reinforcement Learning: Pong from Pixels

    Andrej Karpathy blog About Hacker's guide to Neural Networks Deep Reinforcement Learning: Pong from ...

  4. 论文笔记之:Asynchronous Methods for Deep Reinforcement Learning

    Asynchronous Methods for Deep Reinforcement Learning ICML 2016 深度强化学习最近被人发现貌似不太稳定,有人提出很多改善的方法,这些方法有很 ...

  5. 深度强化学习:入门(Deep Reinforcement Learning: Scratching the surface)

    RL的方案 两个主要对象:Agent和Environment Agent观察Environment,做出Action,这个Action会对Environment造成一定影响和改变,继而Agent会从新 ...

  6. 深度强化学习(Deep Reinforcement Learning)入门:RL base & DQN-DDPG-A3C introduction

    转自https://zhuanlan.zhihu.com/p/25239682 过去的一段时间在深度强化学习领域投入了不少精力,工作中也在应用DRL解决业务问题.子曰:温故而知新,在进一步深入研究和应 ...

  7. (转) Deep Reinforcement Learning: Playing a Racing Game

    Byte Tank Posts Archive Deep Reinforcement Learning: Playing a Racing Game OCT 6TH, 2016 Agent playi ...

  8. Deep Reinforcement Learning with Iterative Shift for Visual Tracking

    Deep Reinforcement Learning with Iterative Shift for Visual Tracking 2019-07-30 14:55:31 Paper: http ...

  9. 论文笔记之:Dueling Network Architectures for Deep Reinforcement Learning

    Dueling Network Architectures for Deep Reinforcement Learning ICML 2016 Best Paper 摘要:本文的贡献点主要是在 DQN ...

随机推荐

  1. Linux内核同步机制之completion

    内核编程中常见的一种模式是,在当前线程之外初始化某个活动,然后等待该活动的结束.这个活动可能是,创建一个新的内核线程或者新的用户空间进程.对一个已有进程的某个请求,或者某种类型的硬件动作,等等.在这种 ...

  2. Java开发环境之ElasticSearch

    查看更多Java开发环境配置,请点击<Java开发环境配置大全> 拾章:ElasticSearch安装教程 1)去官网下载ElasticSearch安装包 http://www.elast ...

  3. MySQL/MariaDB数据库的Galera高可用性集群实战

      MySQL/MariaDB数据库的Galera高可用性集群实战 作者:尹正杰  版权声明:原创作品,谢绝转载!否则将追究法律责任. 一.Galera Cluster概述 1>.什么是Gale ...

  4. Kotlin函数使用综述与显式返回类型分析

    位置参数与具名参数: 继续接着上一次https://www.cnblogs.com/webor2006/p/11498842.html的方法参数学习,再定义一个函数来说明具名参数的问题: 调用一下,先 ...

  5. poj1679The Unique MST(次小生成树模板)

    次小生成树模板,别忘了判定不存在最小生成树的情况 #include <iostream> #include <cstdio> #include <cstring> ...

  6. mysql类似to_char()to_date()函数mysql日期和字符相互转换方法date_f

    mysql 类似to_char() to_date()函数mysql日期和字符相互转换方法 date_format(date,'%Y-%m-%d') -------------->oracle中 ...

  7. Oracle锁表与解锁 对象锁与解锁

    阅读目录 锁表与解锁 查看锁表进程SQL语句 解锁 对象锁与解锁 回到顶部 锁表与解锁 查看锁表进程SQL语句 select * from v$session t1, v$locked_object ...

  8. UVA - 12183 :Top Secret(N^2的循环矩阵乘法)

    pro:N个数排成一圈.一次操作为,每个位置的数+=L*左+R*右,保留x为整数. 问S轮操作后每个位置的值. N<=1000,S<=2^30,x<=9 . sol:不难想到矩阵乘法 ...

  9. test20190925 老L

    100+0+0=100.概率题套路见的太少了,做题策略不是最优的. 排列 给出 n 个二元组,第 i 个二元组为(ai,bi). 将 n 个二元组按照一定顺序排成一列,可以得到一个排列.显然,这样的排 ...

  10. Build Post Office II

    Description Given a 2D grid, each cell is either a wall 2, an house 1 or empty 0 (the number zero, o ...