Multimodal Side-Tuning for Document Classification

TL;DR

This paper introduces a multimodal document classification method using side-tuning, which effectively combines different data sources like text and images, surpassing current accuracy benchmarks.

Contribution

It applies the side-tuning framework to multimodal data, enabling better model adaptation and avoiding issues like model rigidity and catastrophic forgetting.

Findings

Achieves higher accuracy than existing methods

Successfully combines text and image data for classification

Demonstrates effectiveness of side-tuning in multimodal settings

Abstract

In this paper, we propose to exploit the side-tuning framework for multimodal document classification. Side-tuning is a methodology for network adaptation recently introduced to solve some of the problems related to previous approaches. Thanks to this technique it is actually possible to overcome model rigidity and catastrophic forgetting of transfer learning by fine-tuning. The proposed solution uses off-the-shelf deep learning architectures leveraging the side-tuning framework to combine a base model with a tandem of two side networks. We show that side-tuning can be successfully employed also when different data sources are considered, e.g. text and images in document classification. The experimental results show that this approach pushes further the limit for document classification accuracy with respect to the state of the art.