{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2023:L6EMPCJEGBN45NRTUOEWDSLNPR","short_pith_number":"pith:L6EMPCJE","schema_version":"1.0","canonical_sha256":"5f88c78924305bceb633a38961c96d7c587cf6dce53ed0a7b69f20f04f872a4f","source":{"kind":"arxiv","id":"2310.10591","version":1},"attestation_state":"computed","paper":{"title":"Interpreting and Controlling Vision Foundation Models via Text Explanations","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CV","authors_text":"Carl Vondrick, Chengzhi Mao, Haozhe Chen, Junfeng Yang","submitted_at":"2023-10-16T17:12:06Z","abstract_excerpt":"Large-scale pre-trained vision foundation models, such as CLIP, have become de facto backbones for various vision tasks. However, due to their black-box nature, understanding the underlying rules behind these models' predictions and controlling model behaviors have remained open challenges. We present a framework for interpreting vision transformer's latent tokens with natural language. Given a latent token, our framework retains its semantic information to the final layer using transformer's local operations and retrieves the closest text for explanation. Our approach enables understanding of"},"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":"2310.10591","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CV","submitted_at":"2023-10-16T17:12:06Z","cross_cats_sorted":[],"title_canon_sha256":"339160add5af2936053e26ea4f4c48febad27d31c96ffe5e938c10b7ebaf8fef","abstract_canon_sha256":"be1156b9ff7dc3e0003203a2267911d97217cb9929aa7e1dc03777337b1d6a2c"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T07:01:21.982133Z","signature_b64":"85YMXK9r1jLjsYaaaaDGODDO8VeVWsgKoa921OlZ4P2TmsaakNabGW2S/laMUgH+Viu4zX6RAzwVdZ7GlnrQDw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"5f88c78924305bceb633a38961c96d7c587cf6dce53ed0a7b69f20f04f872a4f","last_reissued_at":"2026-07-05T07:01:21.981719Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T07:01:21.981719Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Interpreting and Controlling Vision Foundation Models via Text Explanations","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CV","authors_text":"Carl Vondrick, Chengzhi Mao, Haozhe Chen, Junfeng Yang","submitted_at":"2023-10-16T17:12:06Z","abstract_excerpt":"Large-scale pre-trained vision foundation models, such as CLIP, have become de facto backbones for various vision tasks. However, due to their black-box nature, understanding the underlying rules behind these models' predictions and controlling model behaviors have remained open challenges. We present a framework for interpreting vision transformer's latent tokens with natural language. Given a latent token, our framework retains its semantic information to the final layer using transformer's local operations and retrieves the closest text for explanation. Our approach enables understanding of"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2310.10591","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/2310.10591/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":"2310.10591","created_at":"2026-07-05T07:01:21.981776+00:00"},{"alias_kind":"arxiv_version","alias_value":"2310.10591v1","created_at":"2026-07-05T07:01:21.981776+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2310.10591","created_at":"2026-07-05T07:01:21.981776+00:00"},{"alias_kind":"pith_short_12","alias_value":"L6EMPCJEGBN4","created_at":"2026-07-05T07:01:21.981776+00:00"},{"alias_kind":"pith_short_16","alias_value":"L6EMPCJEGBN45NRT","created_at":"2026-07-05T07:01:21.981776+00:00"},{"alias_kind":"pith_short_8","alias_value":"L6EMPCJE","created_at":"2026-07-05T07:01:21.981776+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":2,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.07451","citing_title":"TEVI: Text-Conditioned Editing of Visual Representations via Sparse Autoencoders for Improved Vision-Language Alignment","ref_index":104,"is_internal_anchor":false},{"citing_arxiv_id":"2511.18359","citing_title":"TRANSPORTER: Transferring Visual Semantics from VLM Manifolds","ref_index":15,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/L6EMPCJEGBN45NRTUOEWDSLNPR","json":"https://pith.science/pith/L6EMPCJEGBN45NRTUOEWDSLNPR.json","graph_json":"https://pith.science/api/pith-number/L6EMPCJEGBN45NRTUOEWDSLNPR/graph.json","events_json":"https://pith.science/api/pith-number/L6EMPCJEGBN45NRTUOEWDSLNPR/events.json","paper":"https://pith.science/paper/L6EMPCJE"},"agent_actions":{"view_html":"https://pith.science/pith/L6EMPCJEGBN45NRTUOEWDSLNPR","download_json":"https://pith.science/pith/L6EMPCJEGBN45NRTUOEWDSLNPR.json","view_paper":"https://pith.science/paper/L6EMPCJE","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2310.10591&json=true","fetch_graph":"https://pith.science/api/pith-number/L6EMPCJEGBN45NRTUOEWDSLNPR/graph.json","fetch_events":"https://pith.science/api/pith-number/L6EMPCJEGBN45NRTUOEWDSLNPR/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/L6EMPCJEGBN45NRTUOEWDSLNPR/action/timestamp_anchor","attest_storage":"https://pith.science/pith/L6EMPCJEGBN45NRTUOEWDSLNPR/action/storage_attestation","attest_author":"https://pith.science/pith/L6EMPCJEGBN45NRTUOEWDSLNPR/action/author_attestation","sign_citation":"https://pith.science/pith/L6EMPCJEGBN45NRTUOEWDSLNPR/action/citation_signature","submit_replication":"https://pith.science/pith/L6EMPCJEGBN45NRTUOEWDSLNPR/action/replication_record"}},"created_at":"2026-07-05T07:01:21.981776+00:00","updated_at":"2026-07-05T07:01:21.981776+00:00"}