Residual networks are an Euler discretization of solutions to Ordinary Differential Equations (ODE). This paper explores a deeper relationship between Transformer and numerical ODE methods. We first show that a residual block of layers in Transformer can be described as a higher-order solution to ODE. Inspired by this, we design a new architecture, {\it ODE Transformer}, which is analogous to the Runge-Kutta method that is well motivated in ODE. As a natural extension to Transformer, ODE Transformer is easy to implement and efficient to use. Experimental results on the large-scale machine translation, abstractive summarization, and grammar error correction tasks demonstrate the high genericity of ODE Transformer. It can gain large improvements in model performance over strong baselines (e.g., 30.77 and 44.11 BLEU scores on the WMT'14 English-German and English-French benchmarks) at a slight cost in inference efficiency.
@article{arxiv.2203.09176,
title = {ODE Transformer: An Ordinary Differential Equation-Inspired Model for Sequence Generation},
author = {Bei Li and Quan Du and Tao Zhou and Yi Jing and Shuhan Zhou and Xin Zeng and Tong Xiao and JingBo Zhu and Xuebo Liu and Min Zhang},
journal= {arXiv preprint arXiv:2203.09176},
year = {2022}
}
Comments
Long paper accepted by ACL2022 main conference. arXiv admin note: substantial text overlap with arXiv:2104.02308