A Theoretically Grounded Application of Dropout in Recurrent Neural Networks
This work addresses overfitting issues in RNNs for practitioners in natural language processing, offering a novel method that improves performance on specific tasks, though it is incremental in extending existing variational tools.
The paper tackled the problem of overfitting in recurrent neural networks (RNNs) by applying a theoretically grounded dropout technique based on variational inference to LSTM and GRU models, resulting in improved performance on language modeling and sentiment analysis tasks, including a state-of-the-art test perplexity of 73.4 on the Penn Treebank.
Recurrent neural networks (RNNs) stand at the forefront of many recent developments in deep learning. Yet a major difficulty with these models is their tendency to overfit, with dropout shown to fail when applied to recurrent layers. Recent results at the intersection of Bayesian modelling and deep learning offer a Bayesian interpretation of common deep learning techniques such as dropout. This grounding of dropout in approximate Bayesian inference suggests an extension of the theoretical results, offering insights into the use of dropout with RNN models. We apply this new variational inference based dropout technique in LSTM and GRU models, assessing it on language modelling and sentiment analysis tasks. The new approach outperforms existing techniques, and to the best of our knowledge improves on the single model state-of-the-art in language modelling with the Penn Treebank (73.4 test perplexity). This extends our arsenal of variational tools in deep learning.