Actor and critc

ballet-545291_1280.jpg

Actor and critic is one of the state of art algorithms in deep reinforcement learning.

1. Brain

   1.1 Model

  We need two models for actor and critic. Fully connected layer, convolutional neural network and recurrent neural network can be candidates for each of them. Like AlphaGo  Zero,  we can combine actor and critic model into one same model. 

   1.2 Loss functions (Object functions)

   1.3 Optimization

 

_TOSHISTATS20180201.png

2. Agent

   2.1 Choice of action 

 

3. Environment

   3.1 Forward step

   3.2 Provide observation and reward to agent