Mitigating Modality Prior-Induced Hallucinations in Multimodal Large Language Models via Deciphering Attention Causality

TL;DR

This paper introduces CausalMM, a causal inference framework that reduces modality prior-induced hallucinations in multimodal large language models by applying structural causal modeling and counterfactual reasoning to improve input-output alignment.

Contribution

It presents a novel causal inference approach using structural causal modeling to mitigate biases from modality priors in MLLMs, enhancing their alignment and reducing hallucinations.

Findings

Achieved up to 65.3% score improvement on VLind-Bench indicators.

Improved MME Benchmark scores by 164 points.

Validated effectiveness through extensive experiments.

Abstract

Multimodal Large Language Models (MLLMs) have emerged as a central focus in both industry and academia, but often suffer from biases introduced by visual and language priors, which can lead to multimodal hallucination. These biases arise from the visual encoder and the Large Language Model (LLM) backbone, affecting the attention mechanism responsible for aligning multimodal inputs. Existing decoding-based mitigation methods focus on statistical correlations and overlook the causal relationships between attention mechanisms and model output, limiting their effectiveness in addressing these biases. To tackle this issue, we propose a causal inference framework termed CausalMM that applies structural causal modeling to MLLMs, treating modality priors as a confounder between attention mechanisms and output. Specifically, by employing backdoor adjustment and counterfactual reasoning at both the visual and language attention levels, our method mitigates the negative effects of modality priors and enhances the alignment of MLLM's inputs and outputs, with a maximum score improvement of 65.3% on 6 VLind-Bench indicators and 164 points on MME Benchmark compared to conventional methods. Extensive experiments validate the effectiveness of our approach while being a plug-and-play solution. Our code is available at: https://github.com/The-Martyr/CausalMM