Sequential Neural Models with Stochastic Layers
This addresses uncertainty propagation in sequential data modeling for applications like speech and music processing, representing a novel method for a known bottleneck rather than incremental.
The paper tackled the problem of efficiently propagating uncertainty in latent state representations for recurrent neural networks by introducing stochastic recurrent neural networks that combine deterministic RNNs with state space models. The approach achieved state-of-the-art results on Blizzard and TIMIT speech datasets with large margins and comparable performance on polyphonic music modeling.
How can we efficiently propagate uncertainty in a latent state representation with recurrent neural networks? This paper introduces stochastic recurrent neural networks which glue a deterministic recurrent neural network and a state space model together to form a stochastic and sequential neural generative model. The clear separation of deterministic and stochastic layers allows a structured variational inference network to track the factorization of the model's posterior distribution. By retaining both the nonlinear recursive structure of a recurrent neural network and averaging over the uncertainty in a latent path, like a state space model, we improve the state of the art results on the Blizzard and TIMIT speech modeling data sets by a large margin, while achieving comparable performances to competing methods on polyphonic music modeling.