English

An Empirical Comparison of Neural Architectures for Reinforcement Learning in Partially Observable Environments

Neural and Evolutionary Computing 2015-12-18 v1 Artificial Intelligence Machine Learning

Abstract

This paper explores the performance of fitted neural Q iteration for reinforcement learning in several partially observable environments, using three recurrent neural network architectures: Long Short-Term Memory, Gated Recurrent Unit and MUT1, a recurrent neural architecture evolved from a pool of several thousands candidate architectures. A variant of fitted Q iteration, based on Advantage values instead of Q values, is also explored. The results show that GRU performs significantly better than LSTM and MUT1 for most of the problems considered, requiring less training episodes and less CPU time before learning a very good policy. Advantage learning also tends to produce better results.

Keywords

Cite

@article{arxiv.1512.05509,
  title  = {An Empirical Comparison of Neural Architectures for Reinforcement Learning in Partially Observable Environments},
  author = {Denis Steckelmacher and Peter Vrancx},
  journal= {arXiv preprint arXiv:1512.05509},
  year   = {2015}
}

Comments

Presented at the 27th Benelux Conference on Artificial Intelligence