Semi-Amortized Variational Autoencoders. Kim, Y.; Wiseman, S.; Miller, A. C.; Sontag, D.; and Rush, A. M. In Proceedings of the 35th International Conference on Machine Learning (ICML), 2018.
Semi-Amortized Variational Autoencoders [pdf]Paper  abstract   bibtex   
Amortized variational inference (AVI) replaces instance-specific local inference with a global inference network. While AVI has enabled efficient training of deep generative models such as variational autoencoders (VAE), recent empirical work suggests that inference networks can produce suboptimal variational parameters. We propose a hybrid approach, to use AVI to initialize the variational parameters and run stochastic variational inference (SVI) to refine them. Crucially, the local SVI procedure is itself differentiable, so the inference network and generative model can be trained end-to-end with gradient-based optimization. This semi-amortized approach enables the use of rich generative models without experiencing the posterior-collapse phenomenon common in training VAEs for problems like text generation. Experiments show this approach outperforms strong autoregressive and variational baselines on standard text and image datasets.
@inproceedings{KimEtAl_icml18,
  author    = {Yoon Kim and Sam Wiseman and Andrew C. Miller and David Sontag and Alexander M. Rush},
  title = {Semi-Amortized Variational Autoencoders},
  booktitle = {Proceedings of the 35th International Conference on Machine Learning ({ICML})},
  year = 2018,
  keywords = {Machine learning, Unsupervised learning, Deep learning, Approximate inference in graphical models},
  url_Paper = {https://arxiv.org/pdf/1802.02550.pdf},
  abstract = {Amortized variational inference (AVI) replaces instance-specific local inference with a global inference network. While AVI has enabled efficient training of deep generative models such as variational autoencoders (VAE), recent empirical work suggests that inference networks can produce suboptimal variational parameters. We propose a hybrid approach, to use AVI to initialize the variational parameters and run stochastic variational inference (SVI) to refine them. Crucially, the local SVI procedure is itself differentiable, so the inference network and generative model can be trained end-to-end with gradient-based optimization. This semi-amortized approach enables the use of rich generative models without experiencing the posterior-collapse phenomenon common in training VAEs for problems like text generation. Experiments show this approach outperforms strong autoregressive and variational baselines on standard text and image datasets.}
}
Downloads: 0