Can You Label Less by Using Out-of-Domain Data? Active & Transfer Learning with Few-shot Instructions

TL;DR

This paper introduces ATF, a novel method that leverages pre-trained language models for active transfer learning without fine-tuning, significantly reducing annotation effort in social media toxicity and bias detection.

Contribution

The paper proposes ATF, a fine-tuning-free active transfer learning approach that effectively transfers knowledge from labeled datasets to new domains using pre-trained language models.

Findings

ATF achieves a 10.5% mean AUC gain over no transfer.

Active learning with few samples enhances transfer benefits.

Transfer success depends on the initial performance of PLMs on target tasks.

Abstract

Labeling social-media data for custom dimensions of toxicity and social bias is challenging and labor-intensive. Existing transfer and active learning approaches meant to reduce annotation effort require fine-tuning, which suffers from over-fitting to noise and can cause domain shift with small sample sizes. In this work, we propose a novel Active Transfer Few-shot Instructions (ATF) approach which requires no fine-tuning. ATF leverages the internal linguistic knowledge of pre-trained language models (PLMs) to facilitate the transfer of information from existing pre-labeled datasets (source-domain task) with minimum labeling effort on unlabeled target data (target-domain task). Our strategy can yield positive transfer achieving a mean AUC gain of 10.5% compared to no transfer with a large 22b parameter PLM. We further show that annotation of just a few target-domain samples via active learning can be beneficial for transfer, but the impact diminishes with more annotation effort (26% drop in gain between 100 and 2000 annotated examples). Finally, we find that not all transfer scenarios yield a positive gain, which seems related to the PLMs initial performance on the target-domain task.