{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:6EILIWPKUGDO5QO7VF3MUFGZI5","short_pith_number":"pith:6EILIWPK","schema_version":"1.0","canonical_sha256":"f110b459eaa186eec1dfa976ca14d947584aec570960559b4a59424a167321f3","source":{"kind":"arxiv","id":"2403.11391","version":1},"attestation_state":"computed","paper":{"title":"Investigating the Benefits of Projection Head for Representation Learning","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.CV"],"primary_cat":"cs.LG","authors_text":"Baharan Mirzasoleiman, Eric Gan, Jiayi Ni, Siddharth Joshi, Yihao Xue","submitted_at":"2024-03-18T00:48:58Z","abstract_excerpt":"An effective technique for obtaining high-quality representations is adding a projection head on top of the encoder during training, then discarding it and using the pre-projection representations. Despite its proven practical effectiveness, the reason behind the success of this technique is poorly understood. The pre-projection representations are not directly optimized by the loss function, raising the question: what makes them better? In this work, we provide a rigorous theoretical answer to this question. We start by examining linear models trained with self-supervised contrastive loss. We"},"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":"2403.11391","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2024-03-18T00:48:58Z","cross_cats_sorted":["cs.CV"],"title_canon_sha256":"fd1ff63eec4ce18758eb428a3648ba174aa08ba131a6d7e7b336f911d47f0ab8","abstract_canon_sha256":"fcbb3c36ce1e7d411e1d225a5151f79ac0ca626aec1d697b6e11e16f73716ab5"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T07:57:15.715323Z","signature_b64":"3JHFw4ep1iggjcUK7B55fa0Oi4dQhCzQI6odQgcH0DgjC25xXgwsp8xEYBTBO5+g8KBKt9jK0Z9gqQqqP863Aw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"f110b459eaa186eec1dfa976ca14d947584aec570960559b4a59424a167321f3","last_reissued_at":"2026-07-05T07:57:15.714785Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T07:57:15.714785Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Investigating the Benefits of Projection Head for Representation Learning","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.CV"],"primary_cat":"cs.LG","authors_text":"Baharan Mirzasoleiman, Eric Gan, Jiayi Ni, Siddharth Joshi, Yihao Xue","submitted_at":"2024-03-18T00:48:58Z","abstract_excerpt":"An effective technique for obtaining high-quality representations is adding a projection head on top of the encoder during training, then discarding it and using the pre-projection representations. Despite its proven practical effectiveness, the reason behind the success of this technique is poorly understood. The pre-projection representations are not directly optimized by the loss function, raising the question: what makes them better? In this work, we provide a rigorous theoretical answer to this question. We start by examining linear models trained with self-supervised contrastive loss. We"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2403.11391","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/2403.11391/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":"2403.11391","created_at":"2026-07-05T07:57:15.714859+00:00"},{"alias_kind":"arxiv_version","alias_value":"2403.11391v1","created_at":"2026-07-05T07:57:15.714859+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2403.11391","created_at":"2026-07-05T07:57:15.714859+00:00"},{"alias_kind":"pith_short_12","alias_value":"6EILIWPKUGDO","created_at":"2026-07-05T07:57:15.714859+00:00"},{"alias_kind":"pith_short_16","alias_value":"6EILIWPKUGDO5QO7","created_at":"2026-07-05T07:57:15.714859+00:00"},{"alias_kind":"pith_short_8","alias_value":"6EILIWPK","created_at":"2026-07-05T07:57:15.714859+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":3,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.21590","citing_title":"Radial Basis Function Networks as Projection Heads in Self-Supervised Learning","ref_index":53,"is_internal_anchor":false},{"citing_arxiv_id":"2605.20610","citing_title":"Beyond Routing: Characterising Expert Tuning and Representation in Vision Mixture-of-Experts","ref_index":68,"is_internal_anchor":false},{"citing_arxiv_id":"2605.14654","citing_title":"Beyond Instance-Level Self-Supervision in 3D Multi-Modal Medical Imaging","ref_index":37,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/6EILIWPKUGDO5QO7VF3MUFGZI5","json":"https://pith.science/pith/6EILIWPKUGDO5QO7VF3MUFGZI5.json","graph_json":"https://pith.science/api/pith-number/6EILIWPKUGDO5QO7VF3MUFGZI5/graph.json","events_json":"https://pith.science/api/pith-number/6EILIWPKUGDO5QO7VF3MUFGZI5/events.json","paper":"https://pith.science/paper/6EILIWPK"},"agent_actions":{"view_html":"https://pith.science/pith/6EILIWPKUGDO5QO7VF3MUFGZI5","download_json":"https://pith.science/pith/6EILIWPKUGDO5QO7VF3MUFGZI5.json","view_paper":"https://pith.science/paper/6EILIWPK","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2403.11391&json=true","fetch_graph":"https://pith.science/api/pith-number/6EILIWPKUGDO5QO7VF3MUFGZI5/graph.json","fetch_events":"https://pith.science/api/pith-number/6EILIWPKUGDO5QO7VF3MUFGZI5/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/6EILIWPKUGDO5QO7VF3MUFGZI5/action/timestamp_anchor","attest_storage":"https://pith.science/pith/6EILIWPKUGDO5QO7VF3MUFGZI5/action/storage_attestation","attest_author":"https://pith.science/pith/6EILIWPKUGDO5QO7VF3MUFGZI5/action/author_attestation","sign_citation":"https://pith.science/pith/6EILIWPKUGDO5QO7VF3MUFGZI5/action/citation_signature","submit_replication":"https://pith.science/pith/6EILIWPKUGDO5QO7VF3MUFGZI5/action/replication_record"}},"created_at":"2026-07-05T07:57:15.714859+00:00","updated_at":"2026-07-05T07:57:15.714859+00:00"}