{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2023:NOOQFNEFPTNXPN75UIILOY3MRF","short_pith_number":"pith:NOOQFNEF","schema_version":"1.0","canonical_sha256":"6b9d02b4857cdb77b7fda210b7636c8945e190e2fd91b4f398911658a1b23e69","source":{"kind":"arxiv","id":"2305.09828","version":1},"attestation_state":"computed","paper":{"title":"Mimetic Initialization of Self-Attention Layers","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI","cs.LG","stat.ML"],"primary_cat":"cs.CV","authors_text":"Asher Trockman, J. Zico Kolter","submitted_at":"2023-05-16T22:12:25Z","abstract_excerpt":"It is notoriously difficult to train Transformers on small datasets; typically, large pre-trained models are instead used as the starting point. We explore the weights of such pre-trained Transformers (particularly for vision) to attempt to find reasons for this discrepancy. Surprisingly, we find that simply initializing the weights of self-attention layers so that they \"look\" more like their pre-trained counterparts allows us to train vanilla Transformers faster and to higher final accuracies, particularly on vision tasks such as CIFAR-10 and ImageNet classification, where we see gains in acc"},"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":"2305.09828","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CV","submitted_at":"2023-05-16T22:12:25Z","cross_cats_sorted":["cs.AI","cs.LG","stat.ML"],"title_canon_sha256":"85d52a10bb18ea16e3e40e8a4901d7730ced87b8c20664c8e1091339f6a85d78","abstract_canon_sha256":"ffbf20439883f3c42057cfd2d4c29c9a9271efa0cbff84ddca5704b30792b1d9"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T06:10:54.297962Z","signature_b64":"E/6tqo+jjyPi51SE1kS3GDl8+3SKDGal+7yko3WVb4/hUQQjhfuvwktskBhsRhtZDf5/MAh5iZdJ8FEyKmquCg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"6b9d02b4857cdb77b7fda210b7636c8945e190e2fd91b4f398911658a1b23e69","last_reissued_at":"2026-07-05T06:10:54.297549Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T06:10:54.297549Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Mimetic Initialization of Self-Attention Layers","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI","cs.LG","stat.ML"],"primary_cat":"cs.CV","authors_text":"Asher Trockman, J. Zico Kolter","submitted_at":"2023-05-16T22:12:25Z","abstract_excerpt":"It is notoriously difficult to train Transformers on small datasets; typically, large pre-trained models are instead used as the starting point. We explore the weights of such pre-trained Transformers (particularly for vision) to attempt to find reasons for this discrepancy. Surprisingly, we find that simply initializing the weights of self-attention layers so that they \"look\" more like their pre-trained counterparts allows us to train vanilla Transformers faster and to higher final accuracies, particularly on vision tasks such as CIFAR-10 and ImageNet classification, where we see gains in acc"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2305.09828","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/2305.09828/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":"2305.09828","created_at":"2026-07-05T06:10:54.297605+00:00"},{"alias_kind":"arxiv_version","alias_value":"2305.09828v1","created_at":"2026-07-05T06:10:54.297605+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2305.09828","created_at":"2026-07-05T06:10:54.297605+00:00"},{"alias_kind":"pith_short_12","alias_value":"NOOQFNEFPTNX","created_at":"2026-07-05T06:10:54.297605+00:00"},{"alias_kind":"pith_short_16","alias_value":"NOOQFNEFPTNXPN75","created_at":"2026-07-05T06:10:54.297605+00:00"},{"alias_kind":"pith_short_8","alias_value":"NOOQFNEF","created_at":"2026-07-05T06:10:54.297605+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":2,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.26894","citing_title":"Modeling Local, Global, and Cross-Modal Context in Multimodal 3D MRI","ref_index":49,"is_internal_anchor":false},{"citing_arxiv_id":"2405.13901","citing_title":"Discrete Cosine Transform Based Decorrelated Attention for Vision Transformers","ref_index":34,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/NOOQFNEFPTNXPN75UIILOY3MRF","json":"https://pith.science/pith/NOOQFNEFPTNXPN75UIILOY3MRF.json","graph_json":"https://pith.science/api/pith-number/NOOQFNEFPTNXPN75UIILOY3MRF/graph.json","events_json":"https://pith.science/api/pith-number/NOOQFNEFPTNXPN75UIILOY3MRF/events.json","paper":"https://pith.science/paper/NOOQFNEF"},"agent_actions":{"view_html":"https://pith.science/pith/NOOQFNEFPTNXPN75UIILOY3MRF","download_json":"https://pith.science/pith/NOOQFNEFPTNXPN75UIILOY3MRF.json","view_paper":"https://pith.science/paper/NOOQFNEF","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2305.09828&json=true","fetch_graph":"https://pith.science/api/pith-number/NOOQFNEFPTNXPN75UIILOY3MRF/graph.json","fetch_events":"https://pith.science/api/pith-number/NOOQFNEFPTNXPN75UIILOY3MRF/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/NOOQFNEFPTNXPN75UIILOY3MRF/action/timestamp_anchor","attest_storage":"https://pith.science/pith/NOOQFNEFPTNXPN75UIILOY3MRF/action/storage_attestation","attest_author":"https://pith.science/pith/NOOQFNEFPTNXPN75UIILOY3MRF/action/author_attestation","sign_citation":"https://pith.science/pith/NOOQFNEFPTNXPN75UIILOY3MRF/action/citation_signature","submit_replication":"https://pith.science/pith/NOOQFNEFPTNXPN75UIILOY3MRF/action/replication_record"}},"created_at":"2026-07-05T06:10:54.297605+00:00","updated_at":"2026-07-05T06:10:54.297605+00:00"}