Corpus ID: 220265592

MDP Homomorphic Networks: Group Symmetries in Reinforcement Learning

@article{Pol2020MDPHN,
  title={MDP Homomorphic Networks: Group Symmetries in Reinforcement Learning},
  author={Elise van der Pol and Daniel E. Worrall and H. V. Hoof and Frans A. Oliehoek and M. Welling},
  journal={ArXiv},
  year={2020},
  volume={abs/2006.16908}
}
  • Elise van der Pol, Daniel E. Worrall, +2 authors M. Welling
  • Published 2020
  • Computer Science, Mathematics
  • ArXiv
  • This paper introduces MDP homomorphic networks for deep reinforcement learning. MDP homomorphic networks are neural networks that are equivariant under symmetries in the joint state-action space of an MDP. Current approaches to deep reinforcement learning do not usually exploit knowledge about such structure. By building this prior knowledge into policy and value networks using an equivariance constraint, we can reduce the size of the solution space. We specifically focus on group-structured… CONTINUE READING

    References

    Publications referenced by this paper.
    SHOWING 1-10 OF 36 REFERENCES
    Plannable Approximations to MDP Homomorphisms: Equivariance under Actions
    4
    Learning Steerable Filters for Rotation Equivariant CNNs
    120
    Group Equivariant Convolutional Networks
    482
    General E(2)-Equivariant Steerable CNNs
    25
    Harmonic Networks: Deep Translation and Rotation Equivariance
    268
    Symmetry Learning for Function Approximation in Reinforcement Learning
    7
    Online abstraction with MDP homomorphisms for Deep Learning
    2
    3D Steerable CNNs: Learning Rotationally Equivariant Features in Volumetric Data
    85
    A General Theory of Equivariant CNNs on Homogeneous Spaces
    52
    Deep Scale-spaces: Equivariance Over Scale
    27