{"about":{"site":"https://codewithpapers.app","non_affiliation":"Code with Papers and Syntology are not affiliated with, endorsed by, or sponsored by Papers with Code, Meta, or the pwc-archive mirror.","licence":"CC BY-SA 4.0","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","attribution":"https://codewithpapers.app/attribution","modified":"archive material modified by Syntology; see the attribution page"},"url":"/paper/interpreting-and-editing-vision-language","title":"Interpreting and Editing Vision-Language Representations to Mitigate Hallucinations","arxiv_id":"2410.02762","date":"2024-10-03","proceeding":null,"authors":["Nick Jiang","Anish Kachinthaya","Suzie Petryk","Yossi Gandelsman"],"abstract":"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.","url_abs":"https://arxiv.org/abs/2410.02762v1","url_pdf":"https://arxiv.org/pdf/2410.02762v1.pdf","source":{"archive":"pwc-archive (Hugging Face), CC BY-SA 4.0","snapshot":"2025-07-28","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","row_kind":"abstracts"},"code_links":[{"paper_slug":"interpreting-and-editing-vision-language","repo_url":"https://github.com/nickjiang2378/vl-interp","is_official":1,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"pytorch","reach":null}],"tasks":[{"task_slug":"zero-shot-segmentation","task_name":"Zero Shot Segmentation"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=2410.02762","mcp":{"get_harvested_code_for_paper":{"arxiv_id":"2410.02762"}},"developers":"https://syntology.ai/developers","read_at":"2026-09-24T18:15:14+00:00","read_at_is":"when the build read Syntology's graph, not when any sample ran","claim":"Per-sample execution status on synthesized fixtures; not a correctness claim about the paper. Samples come from repositories linked to the paper, official or community; repo_kind says which.","repos":[{"provenance":"external:paperswithcode_snapshot_2025-07-28","url":"https://github.com/nickjiang2378/vl-interp","reach":null}],"summary":{"ran":2,"unverified":1},"by_repo_kind":{"official":{"samples":3,"ran":2,"repositories":1}},"repo_kind_vocabulary":{"official":"The archive marks this repository official for the paper","named_in_paper":"The archive records that the paper mentions this repository; it is not marked official","listed":"In the archive's code links for this paper, not marked official and not recorded as mentioned in the paper","found_in_text":"Syntology found this repository in the paper's own text; whether it is the authors' implementation is not asserted","community":"Not in the archive's code links for this paper; a community repository Syntology harvested"},"n_pointer_only_for_licence":0,"samples":[{"code_sha256_prefix":"f04ac0937ce8c5b5","entry":"projection","repo":"nickjiang2378/vl-interp","repo_kind":"official","path":"methods/algorithms.py","file_url":"https://github.com/nickjiang2378/vl-interp/blob/HEAD/methods/algorithms.py","link_basis":"first_harvest_node","language":"python","status":"ran","verification_level":1,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":true,"licence":"CC0-1.0","inline_ok":true,"mcp_get_code":{"code_sha256":"f04ac0937ce8c5b5"}},{"code_sha256_prefix":"d1318c0343cadcb4","entry":"subtract_projection","repo":"nickjiang2378/vl-interp","repo_kind":"official","path":"methods/algorithms.py","file_url":"https://github.com/nickjiang2378/vl-interp/blob/HEAD/methods/algorithms.py","link_basis":"first_harvest_node","language":"python","status":"ran","verification_level":1,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":true,"licence":"CC0-1.0","inline_ok":true,"mcp_get_code":{"code_sha256":"d1318c0343cadcb4"}},{"code_sha256_prefix":"162e38973352a1d4","entry":"subtract_projections","repo":"nickjiang2378/vl-interp","repo_kind":"official","path":"methods/algorithms.py","file_url":"https://github.com/nickjiang2378/vl-interp/blob/HEAD/methods/algorithms.py","link_basis":"first_harvest_node","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"CC0-1.0","inline_ok":true,"mcp_get_code":{"code_sha256":"162e38973352a1d4"}}]},"arxiv_metadata":null,"syntology_extracted_results":null}