ACKTR, or Actor Critic with Kronecker-factored Trust Region, is an actor-critic method for reinforcement learning that applies trust region optimization using a recently proposed Kronecker-factored approximation to the curvature. The method extends the framework of natural policy gradient and optimizes both the actor and the critic using Kronecker-factored approximate curvature (K-FAC) with trust region.
Source: Scalable trust-region method for deep reinforcement learning using Kronecker-factored approximationPaper | Code | Results | Date | Stars |
---|
Task | Papers | Share |
---|---|---|
Reinforcement Learning (RL) | 2 | 33.33% |
Imitation Learning | 1 | 16.67% |
OpenAI Gym | 1 | 16.67% |
Atari Games | 1 | 16.67% |
Continuous Control | 1 | 16.67% |
Component | Type |
|
---|---|---|
Convolution
|
Convolutions | |
Dense Connections
|
Feedforward Networks | |
ELU
|
Activation Functions | (optional) |
Entropy Regularization
|
Regularization | |
ReLU
|
Activation Functions | (optional) |
Softmax
|
Output Functions | (optional) |
Tanh Activation
|
Activation Functions | (optional) |