Towards Developing a Multilingual and Code-Mixed Visual Question Answering System by Knowledge Distillation

TL;DR

This paper introduces a knowledge distillation method to develop a multilingual and code-mixed visual question answering system, leveraging intermediate layer learning and a new multilingual dataset, to improve performance across eleven diverse languages.

Contribution

It proposes a novel multi-layer knowledge distillation approach and creates a large-scale multilingual, code-mixed VQA dataset, enhancing multilingual VQA capabilities.

Findings

The proposed model outperforms existing pre-trained models on eleven language setups.

Multi-layer distillation improves knowledge transfer from teacher to student.

The new dataset enables effective training and evaluation of multilingual VQA systems.

Abstract

Pre-trained language-vision models have shown remarkable performance on the visual question answering (VQA) task. However, most pre-trained models are trained by only considering monolingual learning, especially the resource-rich language like English. Training such models for multilingual setups demand high computing resources and multilingual language-vision dataset which hinders their application in practice. To alleviate these challenges, we propose a knowledge distillation approach to extend an English language-vision model (teacher) into an equally effective multilingual and code-mixed model (student). Unlike the existing knowledge distillation methods, which only use the output from the last layer of the teacher network for distillation, our student model learns and imitates the teacher from multiple intermediate layers (language and vision encoders) with appropriately designed distillation objectives for incremental knowledge extraction. We also create the large-scale multilingual and code-mixed VQA dataset in eleven different language setups considering the multiple Indian and European languages. Experimental results and in-depth analysis show the effectiveness of the proposed VQA model over the pre-trained language-vision models on eleven diverse language setups.