mlsanityData to 14 Oct 2026

Interpreting and Editing Vision-Language Representations to Mitigate Hallucinations

arXiv 2410.02762 ↗First posted 3 Oct 2024cs.CV cs.LGcited by 104

On the map: Multimodal LLMs / vision-language models → Hallucination mitigation in LVLMs

also in 1 more
Abstract (arXiv)

We investigate the internal representations of vision-language models (VLMs) to address hallucinations, a persistent challenge despite advances in model size and training. We project VLMs' internal image representations to their language vocabulary and observe more confident output probabilities on real objects than hallucinated objects. We additionally use these output probabilities to spatially localize real objects. Building on this approach, we introduce a knowledge erasure algorithm that removes hallucinations by linearly orthogonalizing image features with respect to hallucinated object features. We show that targeted edits to a model's latent representations can reduce hallucinations by up to 25.7% on the COCO2014 dataset while preserving performance. Our findings demonstrate how a deeper understanding of VLMs' latent representations can enhance reliability and enable novel capabilities, such as zero-shot segmentation.

Key facts. Cited by 104 papers in the map (data through 14 Oct 2026). Placed in Hallucination mitigation in LVLMs, one of 16 directions in Multimodal LLMs / vision-language models. Builds on 39 of its 53 references that are in the map. First posted on arXiv 3 Oct 2024.

Builds on

39 of 53 references.

Closest ancestors · 5

Same direction, newest first

  1. 2024Reflective Instruction Tuning: Mitigating Hallucinations in Large Vision-Language Models16 Jul
  2. 2023OPERA: Alleviating Hallucination in Multi-Modal Large Language Models via Over-Trust Penalty and Retrospection-Allocation29 Nov
  3. 2023Woodpecker: Hallucination Correction for Multimodal Large Language Models24 Oct
  4. 2023Analyzing and Mitigating Object Hallucination in Large Vision-Language Models1 Oct
  5. 2023Evaluating Object Hallucination in Large Vision-Language Models17 May

Other references · 34

  1. 2021Learning Transferable Visual Models From Natural Language Supervisioncited by 33,744
  2. 2014Microsoft COCO: Common Objects in Contextcited by 17,608
  3. 2023BLIP-2: Bootstrapping Language-Image Pre-training with Frozen Image Encoders and Large Language Modelscited by 6,137
  4. 2023Improved Baselines with Visual Instruction Tuningcited by 4,418
  5. 2022Locating and Editing Factual Associations in GPTcited by 2,691
  6. 2022Survey of Hallucination in Natural Language Generationcited by 2,075
  7. 2023Sparse Autoencoders Find Highly Interpretable Features in Language Modelscited by 1,420
  8. 2018RISE: Randomized Input Sampling for Explanation of Black-box Modelscited by 838
  9. 2023Towards Automated Circuit Discovery for Mechanistic Interpretabilitycited by 724
  10. 2018Object Hallucination in Image Captioningcited by 669
  11. 2023The Internal State of an LLM Knows When It's Lyingcited by 659
  12. 2023Eliciting Latent Predictions from Transformers with the Tuned Lenscited by 623
22 more
  1. 2022Emergent World Representations: Exploring a Sequence Model Trained on a Synthetic Taskcited by 431
  2. 2024INSIDE: LLMs' Internal States Retain the Power of Hallucination Detectioncited by 323
  3. 2021Generic Attention-model Explainability for Interpreting Bi-Modal and Encoder-Decoder Transformerscited by 320
  4. 2020Understanding the Role of Individual Units in a Deep Neural Networkcited by 255
  5. 2024Patchscopes: A Unifying Framework for Inspecting Hidden Representations of Language Modelscited by 235
  6. 2023A Stitch in Time Saves Nine: Detecting and Mitigating Hallucinations of LLMs by Validating Low-Confidence Generationcited by 206
  7. 2023Interpreting CLIP's Image Representation via Text-Based Decompositioncited by 196
  8. 2024TruthX: Alleviating Hallucinations by Editing Large Language Models in Truthful Spacecited by 91
  9. 2024Unsupervised Real-Time Hallucination Detection based on the Internal States of Large Language Modelscited by 86
  10. 2023Overthinking the Truth: Understanding how Language Models Process False Demonstrationscited by 81
  11. 2024A Multimodal Automated Interpretability Agentcited by 63
  12. 2023Towards Vision-Language Mechanistic Interpretability: A Causal Tracing Tool for BLIPcited by 62
  13. 2020A Disentangling Invertible Interpretation Network for Explaining Latent Representationscited by 57
  14. 2023Multimodal Neurons in Pretrained Text-Only Transformerscited by 49
  15. 2023Rosetta Neurons: Mining the Common Units in a Model Zoocited by 46
  16. 2021What if This Modified That? Syntactic Interventions via Counterfactual Embeddingscited by 39
  17. 2024Interpreting the Second-Order Effects of Neurons in CLIPcited by 36
  18. 2024MMNeuron: Discovering Neuron-Level Domain-Specific Interpretation in Multimodal Large Language Modelcited by 26
  19. 2023Interpreting and Controlling Vision Foundation Models via Text Explanationscited by 11
  20. 2024CODIS: Benchmarking Context-Dependent Visual Comprehension for Multimodal Large Language Modelscited by 11
  21. 2023LLM Factoscope: Uncovering LLMs' Factual Discernment through Inner States Analysiscited by 7
  22. 2024Unveiling LLMs: The Evolution of Latent Representations in a Dynamic Knowledge Graphcited by 4

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2024: 5; 2025: 50; 2026: 49.

104 papers.

Sort:

By area:

  1. 2024Devils in Middle Layers of Large Vision-Language Models: Interpreting, Detecting and Mitigating Object Hallucinations via Attention Lenscited by 143
  2. 2024Towards Interpreting Visual Information Processing in Vision-Language Modelscited by 122
  3. 2025Your Large Vision-Language Model Only Needs A Few Attention Heads For Visual Groundingcited by 83
  4. 2025SAeUron: Interpretable Concept Unlearning in Diffusion Models with Sparse Autoencoderscited by 80
  5. 2026Verbalizable Representations Form a Global Workspace in Language Modelscited by 75
  6. 2024What's in the Image? A Deep-Dive into the Vision of Vision Language Modelscited by 68
  7. 2025Steer LLM Latents for Hallucination Detectioncited by 50
  8. 2025Treble Counterfactual VLMs: A Causal Approach to Hallucinationcited by 35
  9. 2025Same Task, Different Circuits: Disentangling Modality-Specific Mechanisms in VLMscited by 35
  10. 2025VideoHallu: Evaluating and Mitigating Multi-modal Hallucinations on Synthetic Video Understandingcited by 32
  11. 2025Pixels Versus Priors: Controlling Knowledge Priors in Vision-Language Models through Visual Counterfactscited by 32
  12. 2025A Survey of Multimodal Hallucination Evaluation and Detectioncited by 27
38 more
  1. 2025From Perception to Cognition: A Survey of Vision-Language Interactive Reasoning in Multimodal Large Language Modelscited by 26
  2. 2025How Visual Representations Map to Language Feature Space in Multimodal LLMscited by 25
  3. 2024Beyond Logit Lens: Contextual Embeddings for Robust Hallucination Detection & Grounding in VLMscited by 22
  4. 2025A Survey on Efficient Vision-Language Modelscited by 19
  5. 2025AudioLens: A Closer Look at Auditory Attribute Perception of Large Audio-Language Modelscited by 18
  6. 2025Forgotten Polygons: Multimodal Large Language Models are Shape-Blindcited by 16
  7. 2025TIDE : Temporal-Aware Sparse Autoencoders for Interpretable Diffusion Transformers in Image Generationcited by 16
  8. 2025Interpreting the Repeated Token Phenomenon in Large Language Modelscited by 16
  9. 2025Where MLLMs Attend and What They Rely On: Explaining Autoregressive Token Generationcited by 16
  10. 2025Hallucinatory Image Tokens: A Training-free EAZY Approach on Detecting and Mitigating Object Hallucinations in LVLMscited by 15
  11. 2025Learning to Steer: Input-dependent Steering for Multimodal LLMscited by 15
  12. 2025Visual Representations inside the Language Modelcited by 15
  13. 2026LatentLens: Revealing Highly Interpretable Visual Tokens in LLMscited by 15
  14. 2026Linear Mechanisms for Spatiotemporal Reasoning in Vision Language Modelscited by 14
  15. 2024Performance Gap in Entity Knowledge Extraction Across Modalities in Vision Language Modelscited by 12
  16. 2025When Seeing Overrides Knowing: Disentangling Knowledge Conflicts in Vision-Language Modelscited by 11
  17. 2025GLSim: Detecting Object Hallucinations in LVLMs via Global-Local Similaritycited by 11
  18. 2026Arbitration Failure, Not Perceptual Blindness: How Vision-Language Models Resolve Visual-Linguistic Conflictscited by 11
  19. 2025SAEmnesia: Erasing Concepts in Diffusion Models with Supervised Sparse Autoencoderscited by 10
  20. 2025LogicQA: Logical Anomaly Detection with Vision Language Model Generated Questionscited by 9
  21. 2025Interpreting Attention Heads for Image-to-Text Information Flow in Large Vision-Language Modelscited by 9
  22. 2025Tell Model Where to Look: Mitigating Hallucinations in MLLMs by Vision-Guided Attentioncited by 9
  23. 2026VIB-Probe: Detecting and Mitigating Hallucinations in Vision-Language Models via Variational Information Bottleneckcited by 8
  24. 2025Mitigating Hallucinations via Inter-Layer Consistency Aggregation in Large Vision-Language Modelscited by 7
  25. 2025Too Late to Recall: Explaining the Two-Hop Problem in Multimodal Knowledge Retrievalcited by 7
  26. 2025Revealing Perception and Generation Dynamics in LVLMs: Mitigating Hallucinations via Validated Dominance Correctioncited by 7
  27. 2026Detecting Misbehaviors of Large Vision-Language Models by Evidential Uncertainty Quantificationcited by 6
  28. 2026Interpretable Debiasing of Vision-Language Models for Social Fairnesscited by 6
  29. 2025Image Tokens Matter: Mitigating Hallucination in Discrete Tokenizer-based Large Vision-Language Models via Latent Editingcited by 5
  30. 2025Seeing the Threat: Vulnerabilities in Vision-Language Models to Adversarial Attackcited by 5
  31. 2025Investigating Mechanisms for In-Context Vision Language Bindingcited by 5
  32. 2025Evidence Recomposition and Predictive Context Residualization for Visual Attribution in Multimodal Large Language Modelscited by 5
  33. 2025SAVE: Sparse Autoencoder-Driven Visual Information Enhancement for Mitigating Object Hallucinationcited by 5
  34. 2026Test-Time Attention Purification for Backdoored Large Vision Language Modelscited by 5
  35. 2025Beyond Cross-Modal Alignment: Measuring and Leveraging Modality Gap in Vision-Language Modelscited by 4
  36. 2025Debiasing CLIP: Interpreting and Correcting Bias in Attention Headscited by 4
  37. 2025Revisit What You See: Revealing Visual Semantics in Vision Tokens to Guide LVLM Decodingcited by 4
  38. 2025Investigating VLM Hallucination from a Cognitive Psychology Perspective: A First Step Toward Interpretation with Intriguing Observationscited by 4

50 most cited of 104 All 104 (JSON)

Similar, not cited

Close in wording, no citation either way

  1. 2024Reducing Hallucinations in Vision-Language Models via Latent Space Steering
  2. 2026Mitigating Hallucinations in Large Vision-Language Models without Performance Degradation
  3. 2026HALP: Detecting Hallucinations in Vision-Language Models without Generating a Single Token
  4. 2024From Pixels to Tokens: Revisiting Object Hallucinations in Large Vision-Language Models
  5. 2025Mitigating Hallucinations in Large Vision-Language Models with Internal Fact-based Contrastive Decoding
  6. 2026NoLan: Mitigating Object Hallucinations in Large Vision-Language Models via Dynamic Suppression of Language Priors
  7. 2025Watch Closely: Mitigating Object Hallucinations in Large Vision-Language Models with Disentangled Decoding
  8. 2025Seeing It or Not? Interpretable Vision-aware Latent Steering to Mitigate Object Hallucinations