BriefGPT.xyz
Jul, 2021
对比对抗训练改进的文本分类
Improved Text Classification via Contrastive Adversarial Training
HTML
PDF
Lin Pan, Chung-Wei Hang, Avirup Sil, Saloni Potdar, Mo Yu
TL;DR
该研究提出了一种正则化Transformer-based编码器fine-tuning用于文本分类任务的通用方法，并通过对干净和对抗性示例的对比学习来实现噪声不变表示，从而得到更好的文本分类效果。
Abstract
We propose a simple and general method to regularize the
fine-tuning
of
transformer-based encoders
for
text classification
tasks. Specific
→