Corpus ID: 219980562

Simple and Effective VAE Training with Calibrated Decoders

@article{Rybkin2020SimpleAE,
  title={Simple and Effective VAE Training with Calibrated Decoders},
  author={Oleh Rybkin and Kostas Daniilidis and Sergey Levine},
  journal={ArXiv},
  year={2020},
  volume={abs/2006.13202}
}
  • Oleh Rybkin, Kostas Daniilidis, Sergey Levine
  • Published 2020
  • Computer Science, Engineering, Mathematics
  • ArXiv
  • Variational autoencoders (VAEs) provide an effective and simple method for modeling complex distributions. However, training VAEs often requires considerable hyperparameter tuning, and often utilizes a heuristic weight on the prior KLdivergence term. In this work, we study how the performance of VAEs can be improved while not requiring the use of this heuristic hyperparameter, by learning calibrated decoders that accurately model the decoding distribution. While in some sense it may seem… CONTINUE READING

    Figures, Tables, and Topics from this paper.

    References

    Publications referenced by this paper.
    SHOWING 1-10 OF 50 REFERENCES

    Fixing a Broken ELBO

    VIEW 7 EXCERPTS
    HIGHLY INFLUENTIAL

    Stochastic Video Generation with a Learned Prior

    VIEW 10 EXCERPTS
    HIGHLY INFLUENTIAL

    Auto-Encoding Variational Bayes

    VIEW 6 EXCERPTS
    HIGHLY INFLUENTIAL

    Deep Learning Face Attributes in the Wild

    VIEW 4 EXCERPTS
    HIGHLY INFLUENTIAL

    Learning Multiple Layers of Features from Tiny Images

    VIEW 4 EXCERPTS
    HIGHLY INFLUENTIAL

    Reading Digits in Natural Images with Unsupervised Feature Learning

    VIEW 4 EXCERPTS
    HIGHLY INFLUENTIAL

    A Neural Representation of Sketch Drawings

    VIEW 1 EXCERPT