Experimental Evaluation of Deep Reinforcement Learning Algorithms

Abstract

Reinforcement learning is an interesting field with regards to its applicability. As such, it has been studied for use in computer vision where it has just recently found its place. The problem it encountered was the high dimensionality of input data in computer vision. The aim of this paper is to provide an overview of the reinforcement learning algorithms for solving the task of reinforcement learning with raw pixels of an image as an input to the algorithm and test their performance on Atari Breakout video game. The comparison of different algorithms will be given and discussed.

Files

Full Paper as PDF

BibTeX Citation

DOI

10.20532/ccvw.2018.0003

https://doi.org/10.20532/ccvw.2018.0003

BibTeX

@InProceedings{10.20532/ccvw.2018.0003,
  author =       {Mrzljak, Nikola and Hrka{\' c}, Tomislav},
  title =        {Experimental Evaluation of Deep Reinforcement
                  Learning Algorithms},
  booktitle =    {Proceedings of the Croatian Compter Vision Workshop,
                  Year 6},
  pages =        {7-12},
  year =         2018,
  editor =       {Lon{\v c}ari{\' c}, Sven and Petkovi{\' c},
                  Tomislav},
  address =      {Zagreb},
  month =        {October},
  organization = {Center of Excellence for Computer Vision},
  publisher =    {University of Zagreb},
  abstract =     {Reinforcement learning is an interesting field with
                  regards to its applicability. As such, it has been
                  studied for use in computer vision where it has just
                  recently found its place. The problem it encountered
                  was the high dimensionality of input data in
                  computer vision. The aim of this paper is to provide
                  an overview of the reinforcement learning algorithms
                  for solving the task of reinforcement learning with
                  raw pixels of an image as an input to the algorithm
                  and test their performance on Atari Breakout video
                  game. The comparison of different algorithms will be
                  given and discussed.},
  doi =          {10.20532/ccvw.2018.0003},
  url =          {https://doi.org/10.20532/ccvw.2018.0003}
}