BriefGPT.xyz
Mar, 2022
对话状态跟踪的继续提示调整
Continual Prompt Tuning for Dialog State Tracking
HTML
PDF
Qi Zhu, Bing Li, Fei Mi, Xiaoyan Zhu, Minlie Huang
TL;DR
本文提出了一种连续的提示调整参数有效的框架,可避免对旧任务的遗忘并在任务之间实现知识转移,验证实验表明该方法在对话状态跟踪领域中具有有效性和高效性。
Abstract
A desirable
dialog system
should be able to continually learn new skills without forgetting old ones, and thereby adapt to new domains or tasks in its life cycle. However, continually training a model often leads to a well-known
→