Alleviating Hallucinations in Large Vision-Language Models through Hallucination-Induced Optimization

2024-05-24

Neural Information Processing Systems

AI Summary

Large Vision-Language Models (LVLMs) commonly suffer from hallucinations—textual outputs inconsistent with the input image. Existing contrastive decoding methods, which rely on global visual uncertainty estimation, fail to precisely localize and suppress hallucinated tokens and may even introduce new hallucinations. To address this, we propose Hallucination-Induced Optimization (HIO), a theory-driven framework featuring: (1) a novel fine-grained hallucination token identification mechanism; and (2) Contrary Bradley–Terry preference modeling coupled with multi-stage logits reweighting to enable targeted contrastive reinforcement between hallucinated and grounded tokens. Unlike prior global uncertainty approaches, HIO operates at the token level, enabling precise hallucination mitigation. Extensive experiments demonstrate that HIO significantly reduces hallucination rates across multiple benchmarks while improving both output faithfulness and cross-modal alignment accuracy, consistently outperforming state-of-the-art methods.

Technology Category

• Computer Vision: Large Vision Models
• Machine Learning: Large Multimodal Models (LMMs)
• Natural Language Processing: Language Grounding & Multi-modal NLP

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation
• Economics, Online Markets and Human Computation: Cost models of using LLMs in production systems

Abstract

Although Large Visual Language Models (LVLMs) have demonstrated exceptional abilities in understanding multimodal data, they invariably suffer from hallucinations, leading to a disconnect between the generated text and the corresponding images. Almost all current visual contrastive decoding methods attempt to mitigate these hallucinations by introducing visual uncertainty information that appropriately widens the contrastive logits gap between hallucinatory and targeted ones. However, due to uncontrollable nature of the global visual uncertainty, they struggle to precisely induce the hallucinatory tokens, which severely limits their effectiveness in mitigating hallucinations and may even lead to the generation of undesired hallucinations. To tackle this issue, we conducted the theoretical analysis to promote the effectiveness of contrast decoding. Building on this insight, we introduce a novel optimization strategy named Hallucination-Induced Optimization (HIO). This strategy seeks to amplify the contrast between hallucinatory and targeted tokens relying on a fine-tuned theoretical preference model (i.e., Contrary Bradley-Terry Model), thereby facilitating efficient contrast decoding to alleviate hallucinations in LVLMs. Extensive experimental research demonstrates that our HIO strategy can effectively reduce hallucinations in LVLMs, outperforming state-of-the-art methods across various benchmarks.

Problem

• Mitigating hallucinations in Large Vision-Language Models (LVLMs)
• Improving contrast decoding for precise hallucination reduction
• Enhancing text-image alignment through Hallucination-Induced Optimization (HIO)

Innovation

• Hallucination-Induced Optimization (HIO) strategy
• Fine-tuned Contrary Bradley-Terry Model
• Amplifies contrast between hallucinatory and targeted tokens