An EM Approach to Non-autoregressive Conditional Sequence Generation

TL;DR

This paper introduces a novel EM-based training framework for non-autoregressive sequence generation that improves accuracy and reduces inference latency by jointly optimizing AR and NAR models.

Contribution

It is the first to apply an EM approach to NAR sequence generation, effectively addressing multi-modality issues and enhancing performance.

Findings

Achieves competitive or better translation quality than existing NAR models.

Significantly reduces inference latency in machine translation tasks.

Demonstrates effectiveness on benchmark datasets.

Abstract

Autoregressive (AR) models have been the dominating approach to conditional sequence generation, but are suffering from the issue of high inference latency. Non-autoregressive (NAR) models have been recently proposed to reduce the latency by generating all output tokens in parallel but could only achieve inferior accuracy compared to their autoregressive counterparts, primarily due to a difficulty in dealing with the multi-modality in sequence generation. This paper proposes a new approach that jointly optimizes both AR and NAR models in a unified Expectation-Maximization (EM) framework. In the E-step, an AR model learns to approximate the regularized posterior of the NAR model. In the M-step, the NAR model is updated on the new posterior and selects the training examples for the next AR model. This iterative process can effectively guide the system to remove the multi-modality in the output sequences. To our knowledge, this is the first EM approach to NAR sequence generation. We evaluate our method on the task of machine translation. Experimental results on benchmark data sets show that the proposed approach achieves competitive, if not better, performance with existing NAR models and significantly reduces the inference latency.