An Exploration of Dropout with RNNs for Natural Language Inference

Gajbhiye, Amit; Jaf, Sardar; Al-Moubayed, Noura; McGough, A. Stephen; Bradley, Steven

Abstract

Dropout is a crucial regularization technique for the Recurrent Neural Network (RNN) models of Natural Language Inference (NLI). However, dropout has not been evaluated for the effectiveness at different layers and dropout rates in NLI models. In this paper, we propose a novel RNN model for NLI and empirically evaluate the effect of applying dropout at different layers in the model. We also investigate the impact of varying dropout rates at these layers. Our empirical evaluation on a large (Stanford Natural Language Inference (SNLI)) and a small (SciTail) dataset suggest that dropout at each feed-forward connection severely affects the model accuracy at increasing dropout rates. We also show that regularizing the embedding layer is efficient for SNLI whereas regularizing the recurrent layer improves the accuracy for SciTail. Our model achieved an accuracy 86.14% on the SNLI dataset and 77.05% on SciTail.

Citation

Gajbhiye, A., Jaf, S., Al-Moubayed, N., McGough, A. S., & Bradley, S. (2018, December). An Exploration of Dropout with RNNs for Natural Language Inference. Presented at ICANN 2018: 27th International Conference on Artificial Neural Networks, Rhodes

• Presentation Conference Type: Conference Paper (published)
• Conference Name: ICANN 2018: 27th International Conference on Artificial Neural Networks
• Acceptance Date: Jul 10, 2018
• Online Publication Date: Oct 1, 2018
• Publication Date: Oct 1, 2018
• Deposit Date: Aug 2, 2018
• Publicly Available Date: Aug 3, 2018
• Print ISSN: 0302-9743
• Publisher: Springer Verlag
• Pages: 157-167
• Series Title: Lecture notes in computer science
• Series Number: 11141
• Book Title: Artificial neural networks and machine learning - ICANN 2018 : 27th international Conference on Artificial Neural Networks, Rhodes, Greece, October 4-7, 2018, proceedings. Part III.
• ISBN: 9783030014230
• DOI: https://doi.org/10.1007/978-3-030-01424-7_16
• Public URL: https://durham-repository.worktribe.com/output/1146132