{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:Y5237MN4WW45A44UCDGIZ3SVU7","short_pith_number":"pith:Y5237MN4","schema_version":"1.0","canonical_sha256":"c775bfb1bcb5b9d0739410cc8cee55a7c2eba58187ad0949004c8e1f61f00b31","source":{"kind":"arxiv","id":"2606.25657","version":1},"attestation_state":"computed","paper":{"title":"Steering Vision-Language Models with Joint Sparse Autoencoders","license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.CV","authors_text":"Hongxu Lin, Hui Li, Huizhen Shu, Wenjie Sun, Xuying Li","submitted_at":"2026-06-24T10:11:04Z","abstract_excerpt":"Sparse Autoencoders (SAEs) have shown promise for analyzing language models, but applying them to vision-language models (VLMs) often yields representations that are difficult to use as controllable cross-modal steering directions. We introduce the Joint Sparse Autoencoder (JSAE), which uses an explicit alignment constraint to jointly factorize sequence-pooled vision and language activations into shared, interpretable image/caption-level features. Applied to LLaVA, JSAE recovers cross-modal features for recognizable concepts (e.g., food and animals). Through bidirectional interventions (additi"},"verification_status":{"content_addressed":true,"pith_receipt":true,"author_attested":false,"weak_author_claims":0,"strong_author_claims":0,"externally_anchored":false,"storage_verified":false,"citation_signatures":0,"replication_records":0,"graph_snapshot":true,"references_resolved":false,"formal_links_present":false},"canonical_record":{"source":{"id":"2606.25657","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","primary_cat":"cs.CV","submitted_at":"2026-06-24T10:11:04Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"94b2fb8af466b30ae1ef3b00ee16c7c96abfde85938df3873a58d4d066c558f7","abstract_canon_sha256":"117c369bc8f8dfbc7d6fee55661ff5eb05233b432a12fc4489e5a72fb4d9b128"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-06-25T01:18:11.964997Z","signature_b64":"qWfp+yZrY6EIFgTAYIBBOQAff6mevEIqUCnO3VfIpm7tZBAAizs6EAQI0hiTOAKOGweZyZslpb42A5YJBojpAQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"c775bfb1bcb5b9d0739410cc8cee55a7c2eba58187ad0949004c8e1f61f00b31","last_reissued_at":"2026-06-25T01:18:11.964543Z","signature_status":"signed_v1","first_computed_at":"2026-06-25T01:18:11.964543Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Steering Vision-Language Models with Joint Sparse Autoencoders","license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.CV","authors_text":"Hongxu Lin, Hui Li, Huizhen Shu, Wenjie Sun, Xuying Li","submitted_at":"2026-06-24T10:11:04Z","abstract_excerpt":"Sparse Autoencoders (SAEs) have shown promise for analyzing language models, but applying them to vision-language models (VLMs) often yields representations that are difficult to use as controllable cross-modal steering directions. We introduce the Joint Sparse Autoencoder (JSAE), which uses an explicit alignment constraint to jointly factorize sequence-pooled vision and language activations into shared, interpretable image/caption-level features. Applied to LLaVA, JSAE recovers cross-modal features for recognizable concepts (e.g., food and animals). Through bidirectional interventions (additi"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2606.25657","kind":"arxiv","version":1},"verdict":{"id":null,"model_set":{},"created_at":null,"strongest_claim":"","one_line_summary":"","pipeline_version":null,"weakest_assumption":"","pith_extraction_headline":""},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2606.25657/integrity.json","findings":[],"available":true,"detectors_run":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938"},"references":{"count":0,"sample":[],"resolved_work":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","internal_anchors":0},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"},"aliases":[{"alias_kind":"arxiv","alias_value":"2606.25657","created_at":"2026-06-25T01:18:11.964601+00:00"},{"alias_kind":"arxiv_version","alias_value":"2606.25657v1","created_at":"2026-06-25T01:18:11.964601+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2606.25657","created_at":"2026-06-25T01:18:11.964601+00:00"},{"alias_kind":"pith_short_12","alias_value":"Y5237MN4WW45","created_at":"2026-06-25T01:18:11.964601+00:00"},{"alias_kind":"pith_short_16","alias_value":"Y5237MN4WW45A44U","created_at":"2026-06-25T01:18:11.964601+00:00"},{"alias_kind":"pith_short_8","alias_value":"Y5237MN4","created_at":"2026-06-25T01:18:11.964601+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":0,"internal_anchor_count":0,"sample":[]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/Y5237MN4WW45A44UCDGIZ3SVU7","json":"https://pith.science/pith/Y5237MN4WW45A44UCDGIZ3SVU7.json","graph_json":"https://pith.science/api/pith-number/Y5237MN4WW45A44UCDGIZ3SVU7/graph.json","events_json":"https://pith.science/api/pith-number/Y5237MN4WW45A44UCDGIZ3SVU7/events.json","paper":"https://pith.science/paper/Y5237MN4"},"agent_actions":{"view_html":"https://pith.science/pith/Y5237MN4WW45A44UCDGIZ3SVU7","download_json":"https://pith.science/pith/Y5237MN4WW45A44UCDGIZ3SVU7.json","view_paper":"https://pith.science/paper/Y5237MN4","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2606.25657&json=true","fetch_graph":"https://pith.science/api/pith-number/Y5237MN4WW45A44UCDGIZ3SVU7/graph.json","fetch_events":"https://pith.science/api/pith-number/Y5237MN4WW45A44UCDGIZ3SVU7/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/Y5237MN4WW45A44UCDGIZ3SVU7/action/timestamp_anchor","attest_storage":"https://pith.science/pith/Y5237MN4WW45A44UCDGIZ3SVU7/action/storage_attestation","attest_author":"https://pith.science/pith/Y5237MN4WW45A44UCDGIZ3SVU7/action/author_attestation","sign_citation":"https://pith.science/pith/Y5237MN4WW45A44UCDGIZ3SVU7/action/citation_signature","submit_replication":"https://pith.science/pith/Y5237MN4WW45A44UCDGIZ3SVU7/action/replication_record"}},"created_at":"2026-06-25T01:18:11.964601+00:00","updated_at":"2026-06-25T01:18:11.964601+00:00"}