Self-Knowledge Distillation in Natural Language Processing

TL;DR

This paper introduces a novel self-knowledge distillation method for NLP that leverages the model’s own soft target probabilities from the embedding space to enhance performance in language modeling and translation tasks.

Contribution

It proposes a new self-knowledge distillation technique that distills multimode information from the embedding space, improving NLP task performance.

Findings

Improved performance in language modeling.

Enhanced neural machine translation results.

Method approximates soft target probabilities for efficiency.

Abstract

Since deep learning became a key player in natural language processing (NLP), many deep learning models have been showing remarkable performances in a variety of NLP tasks, and in some cases, they are even outperforming humans. Such high performance can be explained by efficient knowledge representation of deep learning models. While many methods have been proposed to learn more efficient representation, knowledge distillation from pretrained deep networks suggest that we can use more information from the soft target probability to train other neural networks. In this paper, we propose a new knowledge distillation method self-knowledge distillation, based on the soft target probabilities of the training model itself, where multimode information is distilled from the word embedding space right below the softmax layer. Due to the time complexity, our method approximates the soft target probabilities. In experiments, we applied the proposed method to two different and fundamental NLP tasks: language model and neural machine translation. The experiment results show that our proposed method improves performance on the tasks.