Generative Adversarial Training Can Improve Neural Language Models

TL;DR

This paper introduces a GAN-based adversarial training method that effectively reduces overfitting in neural language models with minimal additional training overhead.

Contribution

It presents a novel regularization technique combining GANs and adversarial training that is more efficient than existing methods like FGSM.

Findings

Reduces overfitting in neural language models

Adds less than 20% training overhead

Improves generalization performance

Abstract

While deep learning in the form of recurrent neural networks (RNNs) has caused a significant improvement in neural language modeling, the fact that they are extremely prone to overfitting is still a mainly unresolved issue. In this paper we propose a regularization method based on generative adversarial networks (GANs) and adversarial training (AT), that can prevent overfitting in neural language models. Unlike common adversarial training methods such as the fast gradient sign method (FGSM) that require a second back-propagation through time, and therefore effectively require at least twice the amount of time for regular training, the overhead of our method does not exceed more than 20% of the training of the baselines.