To Share or not to Share: Predicting Sets of Sources for Model Transfer Learning

TL;DR

This paper introduces a novel method for predicting effective transfer sources in low-resource settings, improving sequence labeling performance by up to 24 F1 points through model similarity and SVM-based predictions.

Contribution

It presents a new approach combining model similarity and SVMs to automatically select sources for transfer learning, addressing limitations of previous ranking methods.

Findings

Predicts promising sources with up to 24 F1 points performance gain.

Shows effectiveness across various domains and tasks.

Demonstrates that source selection improves transfer learning outcomes.

Abstract

In low-resource settings, model transfer can help to overcome a lack of labeled data for many tasks and domains. However, predicting useful transfer sources is a challenging problem, as even the most similar sources might lead to unexpected negative transfer results. Thus, ranking methods based on task and text similarity -- as suggested in prior work -- may not be sufficient to identify promising sources. To tackle this problem, we propose a new approach to automatically determine which and how many sources should be exploited. For this, we study the effects of model transfer on sequence labeling across various domains and tasks and show that our methods based on model similarity and support vector machines are able to predict promising sources, resulting in performance increases of up to 24 F1 points.