ACER

Introduced by Wang et al. in Sample Efficient Actor-Critic with Experience Replay

ACER, or Actor Critic with Experience Replay, is an actor-critic deep reinforcement learning agent with experience replay. It can be seen as an off-policy extension of A3C, where the off-policy estimator is made feasible by:

Using Retrace Q-value estimation.
Using truncated importance sampling with bias correction.
Using a trust region policy optimization method.
Using a stochastic dueling network architecture.

Source: Sample Efficient Actor-Critic with Experience Replay

Read Paper See Code

Papers

Paper	Code	Results	Date	Stars

Tasks

Task	Papers	Share
reinforcement Learning	5	38.46%
Problem Decomposition	2	15.38%
Automatic Speech Recognition	1	7.69%
Speech Recognition	1	7.69%
Face Anti-Spoofing	1	7.69%
Face Recognition	1	7.69%
Spoken Dialogue Systems	1	7.69%
Continuous Control	1	7.69%

Usage Over Time

This feature is experimental; we are continuously improving our matching algorithm.

Components

Component	Type	Add Remove
Convolution	Convolutions
Dense Connections	Feedforward Networks
Entropy Regularization	Regularization
Experience Replay	Replay Memory
ReLU	Activation Functions
Retrace	Value Function Estimation
Softmax	Output Functions
Stochastic Dueling Network	Value Function Estimation
TRPO	Policy Gradient Methods

Categories

Add Remove

Policy Gradient Methods