{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:75ARSMLEIPA5N2TTRGROR4HIRE","short_pith_number":"pith:75ARSMLE","schema_version":"1.0","canonical_sha256":"ff4119316443c1d6ea7389a2e8f0e889210b655147fbe85974e0bbcd6fd76fb2","source":{"kind":"arxiv","id":"2605.03517","version":2},"attestation_state":"computed","paper":{"title":"Understanding Self-Supervised Learning via Latent Distribution Matching","license":"http://creativecommons.org/licenses/by/4.0/","headline":"Self-supervised learning works by matching representations to an assumed latent model while maximizing their entropy to avoid collapse.","cross_cats":["stat.ML"],"primary_cat":"cs.LG","authors_text":"Fabian A Mikulasch, Friedemann Zenke","submitted_at":"2026-05-05T08:53:00Z","abstract_excerpt":"Self-supervised learning (SSL) excels at finding general-purpose latent representations from complex data, yet lacks a unifying theoretical framework that explains the diverse existing methods and guides the design of new ones. We cast SSL as latent distribution matching (LDM): learning representations that maximize their log-probability under an assumed latent model (alignment), while maximizing latent entropy to prevent collapse (uniformity). This view unifies independent component analysis with contrastive, non-contrastive, and predictive SSL methods, including stop gradient approaches. Lev"},"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":"2605.03517","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2026-05-05T08:53:00Z","cross_cats_sorted":["stat.ML"],"title_canon_sha256":"22e38605503f4afc6f2631c9c4851a56388e9d647616cd20ef2234670bf7595d","abstract_canon_sha256":"a7ce7975143800a7b5b8c6c81fc1a0663a8ffce46868a898538ad651689c5e33"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-20T00:05:45.908400Z","signature_b64":"jJvHd+YlVBJjpgYxQGCvPYtIDeiL2Uv6P13rGpQ9E4tUEA5MfSAvVk6g+7iJH1MItw+ezoOzuBVYjEQyL3CtAw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"ff4119316443c1d6ea7389a2e8f0e889210b655147fbe85974e0bbcd6fd76fb2","last_reissued_at":"2026-05-20T00:05:45.907367Z","signature_status":"signed_v1","first_computed_at":"2026-05-20T00:05:45.907367Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Understanding Self-Supervised Learning via Latent Distribution Matching","license":"http://creativecommons.org/licenses/by/4.0/","headline":"Self-supervised learning works by matching representations to an assumed latent model while maximizing their entropy to avoid collapse.","cross_cats":["stat.ML"],"primary_cat":"cs.LG","authors_text":"Fabian A Mikulasch, Friedemann Zenke","submitted_at":"2026-05-05T08:53:00Z","abstract_excerpt":"Self-supervised learning (SSL) excels at finding general-purpose latent representations from complex data, yet lacks a unifying theoretical framework that explains the diverse existing methods and guides the design of new ones. We cast SSL as latent distribution matching (LDM): learning representations that maximize their log-probability under an assumed latent model (alignment), while maximizing latent entropy to prevent collapse (uniformity). This view unifies independent component analysis with contrastive, non-contrastive, and predictive SSL methods, including stop gradient approaches. Lev"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"We cast SSL as latent distribution matching (LDM): learning representations that maximize their log-probability under an assumed latent model (alignment), while maximizing latent entropy to prevent collapse (uniformity). This view unifies independent component analysis with contrastive, non-contrastive, and predictive SSL methods... We further prove that predictive LDM yields identifiable latent representations under mild assumptions, even with nonlinear predictors.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"The existence and suitability of an 'assumed latent model' whose log-probability can be maximized, plus the 'mild assumptions' required for the identifiability proof; without the full derivations it is unclear how restrictive these are or whether they are satisfied by standard SSL objectives.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"Self-supervised learning is cast as latent distribution matching that aligns representations to a model while enforcing uniformity, unifying multiple SSL families and proving identifiability for predictive variants even with nonlinear predictors.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"Self-supervised learning works by matching representations to an assumed latent model while maximizing their entropy to avoid collapse.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"8fe166fd534e11c99720fd0706c74064776e466c19b0884a8817d5b9a30556c6"},"source":{"id":"2605.03517","kind":"arxiv","version":2},"verdict":{"id":"d023f27d-c001-4bf8-99d5-ca81ec816512","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-07T17:02:00.033134Z","strongest_claim":"We cast SSL as latent distribution matching (LDM): learning representations that maximize their log-probability under an assumed latent model (alignment), while maximizing latent entropy to prevent collapse (uniformity). This view unifies independent component analysis with contrastive, non-contrastive, and predictive SSL methods... We further prove that predictive LDM yields identifiable latent representations under mild assumptions, even with nonlinear predictors.","one_line_summary":"Self-supervised learning is cast as latent distribution matching that aligns representations to a model while enforcing uniformity, unifying multiple SSL families and proving identifiability for predictive variants even with nonlinear predictors.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"The existence and suitability of an 'assumed latent model' whose log-probability can be maximized, plus the 'mild assumptions' required for the identifiability proof; without the full derivations it is unclear how restrictive these are or whether they are satisfied by standard SSL objectives.","pith_extraction_headline":"Self-supervised learning works by matching representations to an assumed latent model while maximizing their entropy to avoid collapse."},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2605.03517/integrity.json","findings":[],"available":true,"detectors_run":[{"name":"doi_compliance","ran_at":"2026-05-19T15:17:23.808536Z","status":"completed","version":"1.0.0","findings_count":0}],"snapshot_sha256":"5fb48a1f1cf553c13e95a29491e3e38f9c40780c37e7e09d9fdac09dfdb1f713"},"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":"2605.03517","created_at":"2026-05-20T00:05:45.907492+00:00"},{"alias_kind":"arxiv_version","alias_value":"2605.03517v2","created_at":"2026-05-20T00:05:45.907492+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.03517","created_at":"2026-05-20T00:05:45.907492+00:00"},{"alias_kind":"pith_short_12","alias_value":"75ARSMLEIPA5","created_at":"2026-05-20T00:05:45.907492+00:00"},{"alias_kind":"pith_short_16","alias_value":"75ARSMLEIPA5N2TT","created_at":"2026-05-20T00:05:45.907492+00:00"},{"alias_kind":"pith_short_8","alias_value":"75ARSMLE","created_at":"2026-05-20T00:05:45.907492+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":0,"internal_anchor_count":0,"sample":[]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/75ARSMLEIPA5N2TTRGROR4HIRE","json":"https://pith.science/pith/75ARSMLEIPA5N2TTRGROR4HIRE.json","graph_json":"https://pith.science/api/pith-number/75ARSMLEIPA5N2TTRGROR4HIRE/graph.json","events_json":"https://pith.science/api/pith-number/75ARSMLEIPA5N2TTRGROR4HIRE/events.json","paper":"https://pith.science/paper/75ARSMLE"},"agent_actions":{"view_html":"https://pith.science/pith/75ARSMLEIPA5N2TTRGROR4HIRE","download_json":"https://pith.science/pith/75ARSMLEIPA5N2TTRGROR4HIRE.json","view_paper":"https://pith.science/paper/75ARSMLE","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2605.03517&json=true","fetch_graph":"https://pith.science/api/pith-number/75ARSMLEIPA5N2TTRGROR4HIRE/graph.json","fetch_events":"https://pith.science/api/pith-number/75ARSMLEIPA5N2TTRGROR4HIRE/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/75ARSMLEIPA5N2TTRGROR4HIRE/action/timestamp_anchor","attest_storage":"https://pith.science/pith/75ARSMLEIPA5N2TTRGROR4HIRE/action/storage_attestation","attest_author":"https://pith.science/pith/75ARSMLEIPA5N2TTRGROR4HIRE/action/author_attestation","sign_citation":"https://pith.science/pith/75ARSMLEIPA5N2TTRGROR4HIRE/action/citation_signature","submit_replication":"https://pith.science/pith/75ARSMLEIPA5N2TTRGROR4HIRE/action/replication_record"}},"created_at":"2026-05-20T00:05:45.907492+00:00","updated_at":"2026-05-20T00:05:45.907492+00:00"}