Mitigating Multimodal LLMs Hallucinations via Relevance Propagation at Inference Time

2026-05-03

AI Summary

This work addresses the tendency of multimodal large language models to rely excessively on textual priors during inference, often neglecting perceptual inputs and thereby generating hallucinations. To mitigate this issue, the authors propose LIME, a novel framework that, without modifying model parameters or requiring additional training, introduces Layer-wise Relevance Propagation (LRP) into the inference phase for the first time. LIME quantifies token-level modality contributions and dynamically updates key-value memory representations based on these attributions, thereby enhancing the model’s reliance on visual or auditory inputs. Experiments demonstrate that LIME significantly reduces hallucination rates across multiple multimodal benchmarks while improving semantic alignment and localization capabilities. Importantly, it achieves these gains without compromising generation quality, yielding more rational and interpretable distributions of modality contributions.

Abstract

Multimodal large language models (MLLMs) have revolutionized the landscape of AI, demonstrating impressive capabilities in tackling complex vision and audio-language tasks. However, a critical challenge remains: these models often suffer from hallucinations, generating outputs that diverge from the provided perceptual inputs. This tendency stems from an inherent imbalance in modality utilization during inference, where the dominance of textual tokens undermines the potential of perceptual inputs. As a result, the model frequently resorts to textual language priors at the expense of grounded evidence. To tackle this issue, we propose Learning Inference-time Modality Enhancement (LIME), a training-free framework designed to bolster multimodal grounding by explicitly enhancing modality usage during decoding. LIME leverages Layer-wise Relevance Propagation (LRP) to quantify token-level contributions and defines a relevance-based objective that promotes increased reliance on perceptual inputs. This objective is enforced through inference-time updates to the model's key-value representations, without modifying model parameters or requiring additional training data. We evaluate LIME across multiple multimodal benchmarks in both vision and audio domains, demonstrating consistent reductions in hallucinations and enhanced grounding while preserving generation quality. Further analysis shows that LIME increases modality contribution and produces more localized and semantically aligned relevance patterns.

Problem

• Multimodal LLMs
• Hallucinations
• Modality Imbalance
• Perceptual Grounding
• Inference-time

Innovation

• multimodal hallucination
• inference-time adaptation
• Layer-wise Relevance Propagation
• modality grounding
• training-free method