{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2020:OBFWZHGW7IYO4EJVRDHKF46PTZ","short_pith_number":"pith:OBFWZHGW","schema_version":"1.0","canonical_sha256":"704b6c9cd6fa30ee113588cea2f3cf9e57a815eb63fbf1a0a2d70f9d1f8b2c03","source":{"kind":"arxiv","id":"2005.10419","version":1},"attestation_state":"computed","paper":{"title":"Why distillation helps: a statistical perspective","license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","headline":"","cross_cats":["stat.ML"],"primary_cat":"cs.LG","authors_text":"Aditya Krishna Menon, Ankit Singh Rawat, Sanjiv Kumar, Sashank J. Reddi, Seungyeon Kim","submitted_at":"2020-05-21T01:49:51Z","abstract_excerpt":"Knowledge distillation is a technique for improving the performance of a simple \"student\" model by replacing its one-hot training labels with a distribution over labels obtained from a complex \"teacher\" model. While this simple approach has proven widely effective, a basic question remains unresolved: why does distillation help? In this paper, we present a statistical perspective on distillation which addresses this question, and provides a novel connection to extreme multiclass retrieval techniques. Our core observation is that the teacher seeks to estimate the underlying (Bayes) class-probab"},"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":"2005.10419","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","primary_cat":"cs.LG","submitted_at":"2020-05-21T01:49:51Z","cross_cats_sorted":["stat.ML"],"title_canon_sha256":"ef3b1552cbc3b3f174485d14a97d4c81d250c2ba73bd5f888b6e9a46079aa2d7","abstract_canon_sha256":"7a0ebfa4f30d09f55f0a37687861e6957b5215ed28d6803b1908a65c376ccd47"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T01:04:53.966899Z","signature_b64":"dAc8NHktJ/Bv05LRBuSj6bKvtrgKosT6T/qZsHuycFN6SSZCw80xwZzcu1pWhXN7ziHiRCFBd1xNEZUPVkjMCQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"704b6c9cd6fa30ee113588cea2f3cf9e57a815eb63fbf1a0a2d70f9d1f8b2c03","last_reissued_at":"2026-07-05T01:04:53.966470Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T01:04:53.966470Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Why distillation helps: a statistical perspective","license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","headline":"","cross_cats":["stat.ML"],"primary_cat":"cs.LG","authors_text":"Aditya Krishna Menon, Ankit Singh Rawat, Sanjiv Kumar, Sashank J. Reddi, Seungyeon Kim","submitted_at":"2020-05-21T01:49:51Z","abstract_excerpt":"Knowledge distillation is a technique for improving the performance of a simple \"student\" model by replacing its one-hot training labels with a distribution over labels obtained from a complex \"teacher\" model. While this simple approach has proven widely effective, a basic question remains unresolved: why does distillation help? In this paper, we present a statistical perspective on distillation which addresses this question, and provides a novel connection to extreme multiclass retrieval techniques. Our core observation is that the teacher seeks to estimate the underlying (Bayes) class-probab"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2005.10419","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/2005.10419/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":"2005.10419","created_at":"2026-07-05T01:04:53.966528+00:00"},{"alias_kind":"arxiv_version","alias_value":"2005.10419v1","created_at":"2026-07-05T01:04:53.966528+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2005.10419","created_at":"2026-07-05T01:04:53.966528+00:00"},{"alias_kind":"pith_short_12","alias_value":"OBFWZHGW7IYO","created_at":"2026-07-05T01:04:53.966528+00:00"},{"alias_kind":"pith_short_16","alias_value":"OBFWZHGW7IYO4EJV","created_at":"2026-07-05T01:04:53.966528+00:00"},{"alias_kind":"pith_short_8","alias_value":"OBFWZHGW","created_at":"2026-07-05T01:04:53.966528+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.24747","citing_title":"Scaling Laws for Task-Specific LLM Distillation","ref_index":16,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/OBFWZHGW7IYO4EJVRDHKF46PTZ","json":"https://pith.science/pith/OBFWZHGW7IYO4EJVRDHKF46PTZ.json","graph_json":"https://pith.science/api/pith-number/OBFWZHGW7IYO4EJVRDHKF46PTZ/graph.json","events_json":"https://pith.science/api/pith-number/OBFWZHGW7IYO4EJVRDHKF46PTZ/events.json","paper":"https://pith.science/paper/OBFWZHGW"},"agent_actions":{"view_html":"https://pith.science/pith/OBFWZHGW7IYO4EJVRDHKF46PTZ","download_json":"https://pith.science/pith/OBFWZHGW7IYO4EJVRDHKF46PTZ.json","view_paper":"https://pith.science/paper/OBFWZHGW","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2005.10419&json=true","fetch_graph":"https://pith.science/api/pith-number/OBFWZHGW7IYO4EJVRDHKF46PTZ/graph.json","fetch_events":"https://pith.science/api/pith-number/OBFWZHGW7IYO4EJVRDHKF46PTZ/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/OBFWZHGW7IYO4EJVRDHKF46PTZ/action/timestamp_anchor","attest_storage":"https://pith.science/pith/OBFWZHGW7IYO4EJVRDHKF46PTZ/action/storage_attestation","attest_author":"https://pith.science/pith/OBFWZHGW7IYO4EJVRDHKF46PTZ/action/author_attestation","sign_citation":"https://pith.science/pith/OBFWZHGW7IYO4EJVRDHKF46PTZ/action/citation_signature","submit_replication":"https://pith.science/pith/OBFWZHGW7IYO4EJVRDHKF46PTZ/action/replication_record"}},"created_at":"2026-07-05T01:04:53.966528+00:00","updated_at":"2026-07-05T01:04:53.966528+00:00"}