Prompting ELECTRA: Few-Shot Learning with Discriminative Pre-Trained Models

TL;DR

This paper adapts prompt-based few-shot learning to ELECTRA, a discriminative pre-trained model, demonstrating it outperforms masked language models across various tasks without additional parameters.

Contribution

It introduces a novel prompt-based few-shot learning method for ELECTRA, leveraging its discriminative training to improve performance without extra computational costs.

Findings

ELECTRA outperforms masked language models in few-shot tasks.

The method requires no additional parameters or computation.

ELECTRA learns distributions better aligned with downstream tasks.

Abstract

Pre-trained masked language models successfully perform few-shot learning by formulating downstream tasks as text infilling. However, as a strong alternative in full-shot settings, discriminative pre-trained models like ELECTRA do not fit into the paradigm. In this work, we adapt prompt-based few-shot learning to ELECTRA and show that it outperforms masked language models in a wide range of tasks. ELECTRA is pre-trained to distinguish if a token is generated or original. We naturally extend that to prompt-based few-shot learning by training to score the originality of the target options without introducing new parameters. Our method can be easily adapted to tasks involving multi-token predictions without extra computation overhead. Analysis shows that ELECTRA learns distributions that align better with downstream tasks.