A Survey on Prompt Tuning

TL;DR

This survey comprehensively reviews prompt tuning techniques for language models, categorizing methods, analyzing their designs, and discussing challenges and future research directions.

Contribution

It provides a systematic classification and analysis of prompt tuning methods, highlighting their innovations, advantages, disadvantages, and future challenges.

Findings

Prompt tuning effectively adapts language models with minimal parameter updates.

Various prompt learning methods differ in optimization, encoding, and decomposition strategies.

Challenges include computational efficiency and training stability.

Abstract

This survey reviews prompt tuning, a parameter-efficient approach for adapting language models by prepending trainable continuous vectors while keeping the model frozen. We classify existing approaches into two categories: direct prompt learning and transfer learning. Direct prompt learning methods include: general optimization approaches, encoder-based methods, decomposition strategies, and mixture-of-experts frameworks. Transfer learning methods consist of: general transfer approaches, encoder-based methods, and decomposition strategies. For each method, we analyze method designs, innovations, insights, advantages, and disadvantages, with illustrative visualizations comparing different frameworks. We identify challenges in computational efficiency and training stability, and discuss future directions in improving training robustness and broadening application scope.