{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2023:HDT74OHIYKRD5RRZE2AAREADRO","short_pith_number":"pith:HDT74OHI","schema_version":"1.0","canonical_sha256":"38e7fe38e8c2a23ec63926800890038ba33e82c5fd75742f45cc8b5ab94b84b8","source":{"kind":"arxiv","id":"2307.08122","version":3},"attestation_state":"computed","paper":{"title":"Tangent Transformers for Composition, Privacy and Removal","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Aditya Golatkar, Stefano Soatto, Tian Yu Liu","submitted_at":"2023-07-16T18:31:25Z","abstract_excerpt":"We introduce Tangent Attention Fine-Tuning (TAFT), a method for fine-tuning linearized transformers obtained by computing a First-order Taylor Expansion around a pre-trained initialization. We show that the Jacobian-Vector Product resulting from linearization can be computed efficiently in a single forward pass, reducing training and inference cost to the same order of magnitude as its original non-linear counterpart, while using the same number of parameters. Furthermore, we show that, when applied to various downstream visual classification tasks, the resulting Tangent Transformer fine-tuned"},"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":"2307.08122","kind":"arxiv","version":3},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2023-07-16T18:31:25Z","cross_cats_sorted":[],"title_canon_sha256":"e7ec841ea7b5a7e73061a1912c0038fc19090e63ea7e3268b35a38ec283644d4","abstract_canon_sha256":"e736ca4aa29f6338c7697b68f7a2b10a54f90c4c59144ba58fed7239e69b8311"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T08:19:07.056527Z","signature_b64":"b4mhTkQ00I2UQHH0g5sdU6kNRvcP1XzARQkpk7f+g817oWnhnE76BnsdZgX5ipRNonHxZ68xcjpkPiIEAF5vDg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"38e7fe38e8c2a23ec63926800890038ba33e82c5fd75742f45cc8b5ab94b84b8","last_reissued_at":"2026-07-05T08:19:07.056086Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T08:19:07.056086Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Tangent Transformers for Composition, Privacy and Removal","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Aditya Golatkar, Stefano Soatto, Tian Yu Liu","submitted_at":"2023-07-16T18:31:25Z","abstract_excerpt":"We introduce Tangent Attention Fine-Tuning (TAFT), a method for fine-tuning linearized transformers obtained by computing a First-order Taylor Expansion around a pre-trained initialization. We show that the Jacobian-Vector Product resulting from linearization can be computed efficiently in a single forward pass, reducing training and inference cost to the same order of magnitude as its original non-linear counterpart, while using the same number of parameters. Furthermore, we show that, when applied to various downstream visual classification tasks, the resulting Tangent Transformer fine-tuned"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2307.08122","kind":"arxiv","version":3},"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/2307.08122/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":"2307.08122","created_at":"2026-07-05T08:19:07.056142+00:00"},{"alias_kind":"arxiv_version","alias_value":"2307.08122v3","created_at":"2026-07-05T08:19:07.056142+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2307.08122","created_at":"2026-07-05T08:19:07.056142+00:00"},{"alias_kind":"pith_short_12","alias_value":"HDT74OHIYKRD","created_at":"2026-07-05T08:19:07.056142+00:00"},{"alias_kind":"pith_short_16","alias_value":"HDT74OHIYKRD5RRZ","created_at":"2026-07-05T08:19:07.056142+00:00"},{"alias_kind":"pith_short_8","alias_value":"HDT74OHI","created_at":"2026-07-05T08:19:07.056142+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2506.23104","citing_title":"DC-TTA: Divide-and-Conquer Framework for Test-Time Adaptation of Interactive Segmentation","ref_index":33,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/HDT74OHIYKRD5RRZE2AAREADRO","json":"https://pith.science/pith/HDT74OHIYKRD5RRZE2AAREADRO.json","graph_json":"https://pith.science/api/pith-number/HDT74OHIYKRD5RRZE2AAREADRO/graph.json","events_json":"https://pith.science/api/pith-number/HDT74OHIYKRD5RRZE2AAREADRO/events.json","paper":"https://pith.science/paper/HDT74OHI"},"agent_actions":{"view_html":"https://pith.science/pith/HDT74OHIYKRD5RRZE2AAREADRO","download_json":"https://pith.science/pith/HDT74OHIYKRD5RRZE2AAREADRO.json","view_paper":"https://pith.science/paper/HDT74OHI","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2307.08122&json=true","fetch_graph":"https://pith.science/api/pith-number/HDT74OHIYKRD5RRZE2AAREADRO/graph.json","fetch_events":"https://pith.science/api/pith-number/HDT74OHIYKRD5RRZE2AAREADRO/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/HDT74OHIYKRD5RRZE2AAREADRO/action/timestamp_anchor","attest_storage":"https://pith.science/pith/HDT74OHIYKRD5RRZE2AAREADRO/action/storage_attestation","attest_author":"https://pith.science/pith/HDT74OHIYKRD5RRZE2AAREADRO/action/author_attestation","sign_citation":"https://pith.science/pith/HDT74OHIYKRD5RRZE2AAREADRO/action/citation_signature","submit_replication":"https://pith.science/pith/HDT74OHIYKRD5RRZE2AAREADRO/action/replication_record"}},"created_at":"2026-07-05T08:19:07.056142+00:00","updated_at":"2026-07-05T08:19:07.056142+00:00"}