MuProp: Unbiased Backpropagation for Stochastic Neural Networks
This addresses a key bottleneck in training stochastic networks for complex distributions, offering an incremental improvement over existing gradient estimators.
The paper tackles the problem of training stochastic neural networks with discrete sampling operations, which are not directly compatible with backpropagation, by introducing MuProp, an unbiased gradient estimator that reduces variance using a control variate. The result is consistently good performance on structured output prediction and discrete latent variable modeling tasks.
Deep neural networks are powerful parametric models that can be trained efficiently using the backpropagation algorithm. Stochastic neural networks combine the power of large parametric functions with that of graphical models, which makes it possible to learn very complex distributions. However, as backpropagation is not directly applicable to stochastic networks that include discrete sampling operations within their computational graph, training such networks remains difficult. We present MuProp, an unbiased gradient estimator for stochastic networks, designed to make this task easier. MuProp improves on the likelihood-ratio estimator by reducing its variance using a control variate based on the first-order Taylor expansion of a mean-field network. Crucially, unlike prior attempts at using backpropagation for training stochastic networks, the resulting estimator is unbiased and well behaved. Our experiments on structured output prediction and discrete latent variable modeling demonstrate that MuProp yields consistently good performance across a range of difficult tasks.