Will Multi-modal Data Improves Few-shot Learning?

TL;DR

This paper investigates how adding a text modality to image data enhances few-shot learning, proposing four fusion methods and demonstrating significant accuracy improvements with attention-based fusion.

Contribution

It introduces four fusion techniques for combining image and text features and evaluates their effectiveness on classical few-shot models, highlighting the superior performance of attention-based fusion.

Findings

Attention-based fusion improves accuracy by ~30%.

Adding text modality enhances few-shot learning performance.

Fusion methods outperform baseline models without additional modality.

Abstract

Most few-shot learning models utilize only one modality of data. We would like to investigate qualitatively and quantitatively how much will the model improve if we add an extra modality (i.e. text description of the image), and how it affects the learning procedure. To achieve this goal, we propose four types of fusion method to combine the image feature and text feature. To verify the effectiveness of improvement, we test the fusion methods with two classical few-shot learning models - ProtoNet and MAML, with image feature extractors such as ConvNet and ResNet12. The attention-based fusion method works best, which improves the classification accuracy by a large margin around 30% comparing to the baseline result.