SeqDiffuSeq: Text Diffusion with Encoder-Decoder Transformers

TL;DR

SeqDiffuSeq introduces a novel diffusion-based sequence-to-sequence text generation model using encoder-decoder Transformers, enhancing generation quality and efficiency through self-conditioning and adaptive noise scheduling techniques.

Contribution

It is the first to adapt diffusion models for sequence-to-sequence text generation with innovative noise scheduling and self-conditioning methods.

Findings

Achieves high-quality text generation with improved inference speed.

Demonstrates superior performance over existing models in experiments.

Effectively models denoising across different token positions.

Abstract

Diffusion model, a new generative modelling paradigm, has achieved great success in image, audio, and video generation. However, considering the discrete categorical nature of text, it is not trivial to extend continuous diffusion models to natural language, and text diffusion models are less studied. Sequence-to-sequence text generation is one of the essential natural language processing topics. In this work, we apply diffusion models to approach sequence-to-sequence text generation, and explore whether the superiority generation performance of diffusion model can transfer to natural language domain. We propose SeqDiffuSeq, a text diffusion model for sequence-to-sequence generation. SeqDiffuSeq uses an encoder-decoder Transformers architecture to model denoising function. In order to improve generation quality, SeqDiffuSeq combines the self-conditioning technique and a newly proposed adaptive noise schedule technique. The adaptive noise schedule has the difficulty of denoising evenly distributed across time steps, and considers exclusive noise schedules for tokens at different positional order. Experiment results illustrate the good performance on sequence-to-sequence generation in terms of text quality and inference time.