Corpus ID: 215415964

CURL: Contrastive Unsupervised Representations for Reinforcement Learning

@article{Srinivas2020CURLCU,
  title={CURL: Contrastive Unsupervised Representations for Reinforcement Learning},
  author={A. Srinivas and M. Laskin and P. Abbeel},
  journal={ArXiv},
  year={2020},
  volume={abs/2004.04136}
}
  • A. Srinivas, M. Laskin, P. Abbeel
  • Published 2020
  • Computer Science, Mathematics
  • ArXiv
  • We present CURL: Contrastive Unsupervised Representations for Reinforcement Learning. CURL extracts high-level features from raw pixels using contrastive learning and performs off-policy control on top of the extracted features. CURL outperforms prior pixel-based methods, both model-based and model-free, on complex tasks in the DeepMind Control Suite and Atari Games showing 1.9x and 1.6x performance gains at the 100K environment and interaction steps benchmarks respectively. On the DeepMind… CONTINUE READING
    Debiased Contrastive Learning
    Image Augmentation Is All You Need: Regularizing Deep Reinforcement Learning from Pixels
    18
    Reinforcement Learning with Augmented Data
    13
    Decoupling Representation Learning from Reinforcement Learning
    Learning Invariant Representations for Reinforcement Learning without Reconstruction
    1
    Data-Efficient Reinforcement Learning with Momentum Predictive Representations
    2
    Representation Learning with Video Deep InfoMax
    Automatic Data Augmentation for Generalization in Deep Reinforcement Learning
    1
    Deep Reinforcement and InfoMax Learning
    3

    References

    Publications referenced by this paper.
    SHOWING 1-10 OF 60 REFERENCES
    Momentum Contrast for Unsupervised Visual Representation Learning
    247
    Representation Learning with Contrastive Predictive Coding
    498
    A Simple Framework for Contrastive Learning of Visual Representations
    243
    Unsupervised State Representation Learning in Atari
    39
    Unsupervised Control Through Non-Parametric Discriminative Rewards
    43
    Learning Actionable Representations from Visual Observations
    19
    DARLA: Improving Zero-Shot Transfer in Reinforcement Learning
    172
    Reinforcement Learning with Unsupervised Auxiliary Tasks
    598
    Dueling Network Architectures for Deep Reinforcement Learning
    1136