Implementation in DQN of a rule learning task introduced by Karl Friston for active learning.
The reference to the paper is the following:
Karl J. Friston, Marco Lin, Chris D. Frith, Giovanni Pezzulo, J. Allan Hobson, and Sasha Ondobaka. Active inference, curiosity and insight. Neural Computation, 29:2633โ2683, 2017.
Project report available under report.pdf
A DQN agent was trained over 100 episodes. The resulting learning plot is the following: