Comparing a table and neural network both trained through q-learning

Kidane Mahari Tesfai & Kevin Thor Fleming

Studenteropgave: Bachelorprojekt


The purpose of this project is to implement Q-learning both using a table and a neural
network. Q-learning is an off policy learning method, which learns from the environment
and comes up with a policy, that helps the agent to reach the goal. The first phase of the
report gives a theoretical information of reinforcement learning, then a clear and step
by step implementation of Q learning by hand. The final phase of the report involves
an implementation of Q learning using both methods. The influence of tweaking the
learning parameters is discussed as well. The environments namely FrozenLake-v0 and
FrozenLake8x8-v0 are generated with the Openaigym, which is a toolkit for developing
and comparing reinforcement learning algorithms.

UddannelserDatalogi, (Bachelor/kandidatuddannelse) Bachelor
Udgivelsesdato27 maj 2017
Antal sider51
VejledereCarsten Lunde Petersen