当前位置：网站首页>4. Strategic Learning

4. Strategic Learning

2022-07-08 01:14:00 【C--G】

Policy Gradient with Baseline

Policy Gradient
Insert picture description here

  • BaseLine
    Insert picture description here
    Insert picture description here
  • Monte Carlo Approximation
    Insert picture description here
    Insert picture description here
    Insert picture description here
  • Choices of Baselines
    Choice 1: b=0
    Insert picture description here
    **Choice 2:b is state-value ** Insert picture description here
  • b = VΠ（St）

Insert picture description here
Insert picture description here
Insert picture description here
Insert picture description here

  • Policy Network
    Insert picture description here
  • Value Network
    Insert picture description here
  • Parameter Sharing
    Insert picture description here

Reinforce with Baseline

  • Updating the policy network
    Insert picture description here
    Insert picture description here
    Insert picture description here

Advantage Actor-Critic（A2C）

Insert picture description here
Insert picture description here
Insert picture description here

Reinforce versus A2C

The network structure of the two is almost the same , Different value networks Insert picture description here

A2C with Multi-Step TD Target

one tep
Insert picture description here
Insert picture description here
Multi step
Insert picture description here

Reinforce with Baseline

Insert picture description here

versus

Insert picture description here
Insert picture description here

原网站

版权声明
本文为[C--G]所创，转载请带上原文链接，感谢
https://yzsam.com/2022/189/202207072320355586.html

边栏推荐

猜你喜欢

随机推荐