Memory-Assisted Reinforcement Learning

By Annik Carson (Last updated July 2018)

This code is used to solve reinforcement learning tasks, using a variety of modules. The file structure is as follows:

Environments

Gridworld or OpenAI gym environments which create the tasks to be solved by the RL network

RL Network

Standard RL architecture we develop is an actorcritic network. Can also use Q-learning, etc.

Memory

Episodic caching system used to assist the RL network

Sensory

Networks used to create efficient representations of incoming state information. Can be used to supplement the RL network. These may be modified autoencoders, etc.

Notebooks

Jupyter notebooks used for running code

Data

Storage of data from runs for later analysis

Example Code