Skip to content

A PyTorch re-implementation of "Doubly Reparameterized Gradient Estimators for Monte Carlo Objectives"

Notifications You must be signed in to change notification settings

iffsid/DReG-PyTorch

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

23 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

DReG-PyTorch

A PyTorch re-implementation of "Doubly Reparameterized Gradient Estimators for Monte Carlo Objectives"

What is DReG?

DReG is a variance reduction technique proposed in the paper "Doubly Reparameterized Gradient Estimators for Monte Carlo Objectives" [1]. The work builds on earlier variance reduction techniques for Monte Carlo estimators in latent variable models and reinforcement learning. The central idea extends from observing a score function term in the gradient of the evidence lower bound [2], and replacing it using the estimate given by the reparameterization trck. Since the reparameterization gradient usually has lower variance than the score function estimator/REINFORCE, the overall Monte Carlo estimator has lower variance.

What does DReG do?

  • DReG applied to estimating the gradient of the inference network in a latent variable model trained with amortized inference gives lower variance gradients
  • Addressed concerns about using tight IWAE bounds during training [3]
  • Generalized the gradient estimator in [2] to IWAE without introducing bias
  • Potential extensions and applications in sequence models and model-based RL

Variance Reduction Visualization

TODOs

  • RWS case
  • JVI case

References

[1] Tucker, G., Lawson, D., Gu, S. and Maddison, C.J., 2018. Doubly Reparameterized Gradient Estimators for Monte Carlo Objectives. arXiv preprint arXiv:1810.04152.

[2] Roeder, G., Wu, Y. and Duvenaud, D.K., 2017. Sticking the landing: Simple, lower-variance gradient estimators for variational inference. In Advances in Neural Information Processing Systems (pp. 6925-6934).

[3] Rainforth, T., Kosiorek, A.R., Le, T.A., Maddison, C.J., Igl, M., Wood, F. and Teh, Y.W., 2018. Tighter variational bounds are not necessarily better. arXiv preprint arXiv:1802.04537.

About

A PyTorch re-implementation of "Doubly Reparameterized Gradient Estimators for Monte Carlo Objectives"

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages