Scalable Bayesian Learning of Recurrent Neural Networks for Language Modeling
This addresses the problem of overfitting in RNNs for language modeling, offering a more robust training method for researchers and practitioners, though it is incremental as it builds on existing techniques.
The paper tackled overfitting in recurrent neural networks for language modeling by introducing a Bayesian learning algorithm that uses stochastic gradient Markov Chain Monte Carlo to estimate weight uncertainty, resulting in superior performance over stochastic optimization across various models and applications.
Recurrent neural networks (RNNs) have shown promising performance for language modeling. However, traditional training of RNNs using back-propagation through time often suffers from overfitting. One reason for this is that stochastic optimization (used for large training sets) does not provide good estimates of model uncertainty. This paper leverages recent advances in stochastic gradient Markov Chain Monte Carlo (also appropriate for large training sets) to learn weight uncertainty in RNNs. It yields a principled Bayesian learning algorithm, adding gradient noise during training (enhancing exploration of the model-parameter space) and model averaging when testing. Extensive experiments on various RNN models and across a broad range of applications demonstrate the superiority of the proposed approach over stochastic optimization.