Multimodal Reasoning with Multimodal Knowledge Graph

TL;DR

This paper introduces MR-MKG, a novel method that uses multimodal knowledge graphs to improve large language models' ability to perform multimodal reasoning, reducing hallucinations and enhancing understanding across image and text modalities.

Contribution

The paper presents a new approach leveraging multimodal knowledge graphs and a relation graph attention network to significantly enhance multimodal reasoning in LLMs with minimal additional parameters.

Findings

MR-MKG outperforms previous models on multimodal question answering.

Achieves superior results with only 2.25% of the LLM's parameters.

Constructed a new MMKG-grounded dataset for training and evaluation.

Abstract

Multimodal reasoning with large language models (LLMs) often suffers from hallucinations and the presence of deficient or outdated knowledge within LLMs. Some approaches have sought to mitigate these issues by employing textual knowledge graphs, but their singular modality of knowledge limits comprehensive cross-modal understanding. In this paper, we propose the Multimodal Reasoning with Multimodal Knowledge Graph (MR-MKG) method, which leverages multimodal knowledge graphs (MMKGs) to learn rich and semantic knowledge across modalities, significantly enhancing the multimodal reasoning capabilities of LLMs. In particular, a relation graph attention network is utilized for encoding MMKGs and a cross-modal alignment module is designed for optimizing image-text alignment. A MMKG-grounded dataset is constructed to equip LLMs with initial expertise in multimodal reasoning through pretraining. Remarkably, MR-MKG achieves superior performance while training on only a small fraction of parameters, approximately 2.25% of the LLM's parameter size. Experimental results on multimodal question answering and multimodal analogy reasoning tasks demonstrate that our MR-MKG method outperforms previous state-of-the-art models.