{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:J4C4PLRZ5F3XNVQUSXDQ3VY3HG","short_pith_number":"pith:J4C4PLRZ","schema_version":"1.0","canonical_sha256":"4f05c7ae39e97776d61495c70dd71b3988b72e24a1623adf88d07488b0f17c00","source":{"kind":"arxiv","id":"2404.05729","version":2},"attestation_state":"computed","paper":{"title":"Finding Visual Task Vectors","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CV","authors_text":"Alberto Hojel, Amir Bar, Amir Globerson, Trevor Darrell, Yutong Bai","submitted_at":"2024-04-08T17:59:46Z","abstract_excerpt":"Visual Prompting is a technique for teaching models to perform a visual task via in-context examples, without any additional training. In this work, we analyze the activations of MAE-VQGAN, a recent Visual Prompting model, and find task vectors, activations that encode task-specific information. Equipped with this insight, we demonstrate that it is possible to identify the task vectors and use them to guide the network towards performing different tasks without providing any input-output examples. To find task vectors, we compute the average intermediate activations per task and use the REINFO"},"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":"2404.05729","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CV","submitted_at":"2024-04-08T17:59:46Z","cross_cats_sorted":[],"title_canon_sha256":"1fb73cde907b7fa869d8f6181b02ce5df14e01a2fe26d97d2ce7bbd672f7ef87","abstract_canon_sha256":"a2bbc100d9d4a6694c4a83cde923999e54d384d0ccd6f7cfedf41d95af7fb681"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T09:16:48.070831Z","signature_b64":"fRc+hNn5AhLtra88uFHk2M7Ak5vHkQOkP4RRVuY+pPneyHbM8/EjfWyNtYr7RMzJ+eoGwZm0RJcVyHe9S3OLDw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"4f05c7ae39e97776d61495c70dd71b3988b72e24a1623adf88d07488b0f17c00","last_reissued_at":"2026-07-05T09:16:48.070343Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T09:16:48.070343Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Finding Visual Task Vectors","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CV","authors_text":"Alberto Hojel, Amir Bar, Amir Globerson, Trevor Darrell, Yutong Bai","submitted_at":"2024-04-08T17:59:46Z","abstract_excerpt":"Visual Prompting is a technique for teaching models to perform a visual task via in-context examples, without any additional training. In this work, we analyze the activations of MAE-VQGAN, a recent Visual Prompting model, and find task vectors, activations that encode task-specific information. Equipped with this insight, we demonstrate that it is possible to identify the task vectors and use them to guide the network towards performing different tasks without providing any input-output examples. To find task vectors, we compute the average intermediate activations per task and use the REINFO"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2404.05729","kind":"arxiv","version":2},"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/2404.05729/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":"2404.05729","created_at":"2026-07-05T09:16:48.070401+00:00"},{"alias_kind":"arxiv_version","alias_value":"2404.05729v2","created_at":"2026-07-05T09:16:48.070401+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2404.05729","created_at":"2026-07-05T09:16:48.070401+00:00"},{"alias_kind":"pith_short_12","alias_value":"J4C4PLRZ5F3X","created_at":"2026-07-05T09:16:48.070401+00:00"},{"alias_kind":"pith_short_16","alias_value":"J4C4PLRZ5F3XNVQU","created_at":"2026-07-05T09:16:48.070401+00:00"},{"alias_kind":"pith_short_8","alias_value":"J4C4PLRZ","created_at":"2026-07-05T09:16:48.070401+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.28273","citing_title":"Vision-Default, Prior-Override: Causal Mechanisms of Perception-Knowledge Conflict in Vision-Language Models","ref_index":6,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/J4C4PLRZ5F3XNVQUSXDQ3VY3HG","json":"https://pith.science/pith/J4C4PLRZ5F3XNVQUSXDQ3VY3HG.json","graph_json":"https://pith.science/api/pith-number/J4C4PLRZ5F3XNVQUSXDQ3VY3HG/graph.json","events_json":"https://pith.science/api/pith-number/J4C4PLRZ5F3XNVQUSXDQ3VY3HG/events.json","paper":"https://pith.science/paper/J4C4PLRZ"},"agent_actions":{"view_html":"https://pith.science/pith/J4C4PLRZ5F3XNVQUSXDQ3VY3HG","download_json":"https://pith.science/pith/J4C4PLRZ5F3XNVQUSXDQ3VY3HG.json","view_paper":"https://pith.science/paper/J4C4PLRZ","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2404.05729&json=true","fetch_graph":"https://pith.science/api/pith-number/J4C4PLRZ5F3XNVQUSXDQ3VY3HG/graph.json","fetch_events":"https://pith.science/api/pith-number/J4C4PLRZ5F3XNVQUSXDQ3VY3HG/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/J4C4PLRZ5F3XNVQUSXDQ3VY3HG/action/timestamp_anchor","attest_storage":"https://pith.science/pith/J4C4PLRZ5F3XNVQUSXDQ3VY3HG/action/storage_attestation","attest_author":"https://pith.science/pith/J4C4PLRZ5F3XNVQUSXDQ3VY3HG/action/author_attestation","sign_citation":"https://pith.science/pith/J4C4PLRZ5F3XNVQUSXDQ3VY3HG/action/citation_signature","submit_replication":"https://pith.science/pith/J4C4PLRZ5F3XNVQUSXDQ3VY3HG/action/replication_record"}},"created_at":"2026-07-05T09:16:48.070401+00:00","updated_at":"2026-07-05T09:16:48.070401+00:00"}