Targeting the Attention Heads Behind Object Hallucination in LLaVA

arXiv:2608.24966 · cs.CV, cs.AI · Submitted 2026-08-25 · Read on arXiv

cs.CV, cs.AI

Submitted: 2026-08-25

Updated: 2026-08-25

Comments: 10 pages, 5 figures, 3 tables. Accepted at the Actionable Interpretability Workshop, COLM 2026

License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/

Terminology

• Hallucination — 46× in this paper · explained in 4 episodes
• Grounding — 39× in this paper · explained in 3 episodes
• Interpretability — 17× in this paper · explained in 15 episodes
• Ablation — 9× in this paper · explained in 1 episode
• Adaptation — 6× in this paper · explained in 1 episode
• Conference on Computer Vision and Pattern Recognition — 3× in this paper
• Fine-Tuning — 3× in this paper · explained in 5 episodes
• Robustness — 3× in this paper · explained in 5 episodes
• Empirical Methods in Natural Language Processing — 2× in this paper
• Instruction Tuning — 2× in this paper · explained in 3 episodes
• Visual Contrastive Decoding — 2× in this paper
• Actionability — 1× in this paper · explained in 3 episodes
• In International Conference on Learning Representations — 1× in this paper
• LoRA Adaptation — 1× in this paper · explained in 2 episodes
• Parameter-Efficient Fine-Tuning — 1× in this paper · explained in 3 episodes
• Tokenization — 1× in this paper · explained in 2 episodes
• Transformer — 1× in this paper · explained in 2 episodes

Sources

• QLoRA: Efficient Finetuning of Quantized LLMs
• LoRA: Low-Rank Adaptation of Large Language Models
• Devils in Middle Layers of Large Vision-Language Models: Interpreting, Detecting and Mitigating Object Hallucinations via Attention Lens
• Mitigating Object Hallucinations in Large Vision-Language Models through Visual Contrastive Decoding
• Visual Instruction Tuning
• Improved Baselines with Visual Instruction Tuning
• Paying More Attention to Image: A Training-Free Method for Alleviating Hallucination in LVLMs
• Direct Preference Optimization: Your Language Model is Secretly a Reward Model
• Mitigating Hallucinations in Vision-Language Models through Image-Guided Head Suppression

Related papers

• Loss Knows Best: Detecting Annotation Errors in Videos via Loss Trajectories
• AnchorWeave: World-Consistent Video Generation with Retrieved Local Spatial Memories
• Benchmarking the Robustness of Foundation Models for Mammography under Domain Shift
• MambaX-Net: Dual-Input Mamba-Enhanced Cross-Attention Network for Longitudinal MRI Segmentation
• TeleOCR: Navigating Document Parsing Across Digital and Camera-Captured Documents
• A Survey on Efficient Vision-Language-Action Models