Improving Sequence-to-Sequence Learning via Optimal Transport
Abstract
Sequence-to-sequence models are commonly trained via maximum likelihood estimation (MLE). However, standard MLE training considers a word-level objective, predicting the next word given the previous ground-truth partial sentence. This procedure focuses on modeling local syntactic patterns, and may fail to capture long-range semantic structure. We present a novel solution to alleviate these issues. Our approach imposes global sequence-level guidance via new supervision based on optimal transport, enabling the overall characterization and preservation of semantic features. We further show that this method can be understood as a Wasserstein gradient flow trying to match our model to the ground truth sequence distribution. Extensive experiments are conducted to validate the utility of the proposed approach, showing consistent improvements over a wide variety of NLP tasks, including machine translation, abstractive text summarization, and image captioning.
Cite
@article{arxiv.1901.06283,
title = {Improving Sequence-to-Sequence Learning via Optimal Transport},
author = {Liqun Chen and Yizhe Zhang and Ruiyi Zhang and Chenyang Tao and Zhe Gan and Haichao Zhang and Bai Li and Dinghan Shen and Changyou Chen and Lawrence Carin},
journal= {arXiv preprint arXiv:1901.06283},
year = {2019}
}