Improving Language Modelling with Noise Contrastive Estimation

Liza, Farhana Ferdousi and Grzes, Marek (2018) Improving Language Modelling with Noise Contrastive Estimation. In: The Thirty-Second AAAI Conference on Artificial Intelligence. AAAI Press, Palo Alto, California.. pp. 5277-5284. ISBN 978-1-57735-800-8. (Full text available)

PDF - Author's Accepted Manuscript
Download (238kB) Preview
[img]
Preview

Abstract

Neural language models do not scale well when the vocabulary is large. Noise contrastive estimation (NCE) is a sampling-based method that allows for fast learning with large vocabularies. Although NCE has shown promising performance in neural machine translation, its full potential has not been demonstrated in the language modelling literature. A sufficient investigation of the hyperparameters in the NCE-based neural language models was clearly missing. In this paper, we showed that NCE can be a very successful approach in neural language modelling when the hyperparameters of a neural network are tuned appropriately. We introduced the `search-then-converge' learning rate schedule for NCE and designed a heuristic that specifies how to use this schedule. The impact of the other important hyperparameters, such as the dropout rate and the weight initialisation range, was also demonstrated. Using a popular benchmark, we showed that appropriate tuning of NCE in neural language models outperforms the state-of-the-art single-model methods based on the standard LSTM recurrent neural networks.

Item Type: Conference or workshop item (Proceeding)
Subjects: Q Science > QA Mathematics (inc Computing science) > QA276 Mathematical statistics
Q Science > QA Mathematics (inc Computing science) > QA 76 Software, computer programming, > QA76.87 Neural computers, neural networks
Divisions: Faculties > Sciences > School of Computing
Faculties > Sciences > School of Computing > Computational Intelligence Group
Depositing User: Marek Grzes
Date Deposited: 08 Dec 2017 16:00 UTC
Last Modified: 04 Feb 2019 14:58 UTC
Resource URI: https://kar.kent.ac.uk/id/eprint/65147 (The current URI for this page, for reference purposes)
  • Depositors only (login required):

Downloads

Downloads per month over past year