{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2022:GEOPWP6ESZT63DF4MS555JBVAY","short_pith_number":"pith:GEOPWP6E","schema_version":"1.0","canonical_sha256":"311cfb3fc49667ed8cbc64bbdea4350616cf8e3d9993fd1d041161b1b48a027b","source":{"kind":"arxiv","id":"2206.08491","version":1},"attestation_state":"computed","paper":{"title":"Revisiting Self-Distillation","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Ameya Joshi, Chinmay Hegde, Minh Pham, Minsu Cho","submitted_at":"2022-06-17T00:18:51Z","abstract_excerpt":"Knowledge distillation is the procedure of transferring \"knowledge\" from a large model (the teacher) to a more compact one (the student), often being used in the context of model compression. When both models have the same architecture, this procedure is called self-distillation. Several works have anecdotally shown that a self-distilled student can outperform the teacher on held-out data. In this work, we systematically study self-distillation in a number of settings. We first show that even with a highly accurate teacher, self-distillation allows a student to surpass the teacher in all cases"},"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":"2206.08491","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2022-06-17T00:18:51Z","cross_cats_sorted":[],"title_canon_sha256":"e6a1107fbdef7c87d89b824ade7def599f74f8a50cae40be1e45111c217072db","abstract_canon_sha256":"c4c006be27939480463e46e890effd6a66a0814d01c95d6bcfeb000b7d3f77fe"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T04:32:30.242560Z","signature_b64":"4dMg9vgpmF03NNo/pTyaFnEEk8UOss8TsHy9CyjGL1FlTfFegXnoa/aZ8e5zUaH9UVa8EjP/uvVbAWag9yl4Ag==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"311cfb3fc49667ed8cbc64bbdea4350616cf8e3d9993fd1d041161b1b48a027b","last_reissued_at":"2026-07-05T04:32:30.241936Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T04:32:30.241936Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Revisiting Self-Distillation","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Ameya Joshi, Chinmay Hegde, Minh Pham, Minsu Cho","submitted_at":"2022-06-17T00:18:51Z","abstract_excerpt":"Knowledge distillation is the procedure of transferring \"knowledge\" from a large model (the teacher) to a more compact one (the student), often being used in the context of model compression. When both models have the same architecture, this procedure is called self-distillation. Several works have anecdotally shown that a self-distilled student can outperform the teacher on held-out data. In this work, we systematically study self-distillation in a number of settings. We first show that even with a highly accurate teacher, self-distillation allows a student to surpass the teacher in all cases"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2206.08491","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/2206.08491/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":"2206.08491","created_at":"2026-07-05T04:32:30.242020+00:00"},{"alias_kind":"arxiv_version","alias_value":"2206.08491v1","created_at":"2026-07-05T04:32:30.242020+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2206.08491","created_at":"2026-07-05T04:32:30.242020+00:00"},{"alias_kind":"pith_short_12","alias_value":"GEOPWP6ESZT6","created_at":"2026-07-05T04:32:30.242020+00:00"},{"alias_kind":"pith_short_16","alias_value":"GEOPWP6ESZT63DF4","created_at":"2026-07-05T04:32:30.242020+00:00"},{"alias_kind":"pith_short_8","alias_value":"GEOPWP6E","created_at":"2026-07-05T04:32:30.242020+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":5,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2605.13143","citing_title":"On the Generalization of Knowledge Distillation: An Information-Theoretic View","ref_index":15,"is_internal_anchor":false},{"citing_arxiv_id":"2604.04038","citing_title":"FLAME: Condensing Ensemble Diversity into a Single Network for Efficient Sequential Recommendation","ref_index":38,"is_internal_anchor":false},{"citing_arxiv_id":"2605.11414","citing_title":"Generative Diffusion Prior Distillation for Long-Context Knowledge Transfer","ref_index":9,"is_internal_anchor":false},{"citing_arxiv_id":"2604.07894","citing_title":"TSUBASA: Improving Long-Horizon Personalization via Evolving Memory and Self-Learning with Context Distillation","ref_index":43,"is_internal_anchor":false},{"citing_arxiv_id":"2604.21027","citing_title":"HypEHR: Hyperbolic Modeling of Electronic Health Records for Efficient Question Answering","ref_index":190,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/GEOPWP6ESZT63DF4MS555JBVAY","json":"https://pith.science/pith/GEOPWP6ESZT63DF4MS555JBVAY.json","graph_json":"https://pith.science/api/pith-number/GEOPWP6ESZT63DF4MS555JBVAY/graph.json","events_json":"https://pith.science/api/pith-number/GEOPWP6ESZT63DF4MS555JBVAY/events.json","paper":"https://pith.science/paper/GEOPWP6E"},"agent_actions":{"view_html":"https://pith.science/pith/GEOPWP6ESZT63DF4MS555JBVAY","download_json":"https://pith.science/pith/GEOPWP6ESZT63DF4MS555JBVAY.json","view_paper":"https://pith.science/paper/GEOPWP6E","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2206.08491&json=true","fetch_graph":"https://pith.science/api/pith-number/GEOPWP6ESZT63DF4MS555JBVAY/graph.json","fetch_events":"https://pith.science/api/pith-number/GEOPWP6ESZT63DF4MS555JBVAY/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/GEOPWP6ESZT63DF4MS555JBVAY/action/timestamp_anchor","attest_storage":"https://pith.science/pith/GEOPWP6ESZT63DF4MS555JBVAY/action/storage_attestation","attest_author":"https://pith.science/pith/GEOPWP6ESZT63DF4MS555JBVAY/action/author_attestation","sign_citation":"https://pith.science/pith/GEOPWP6ESZT63DF4MS555JBVAY/action/citation_signature","submit_replication":"https://pith.science/pith/GEOPWP6ESZT63DF4MS555JBVAY/action/replication_record"}},"created_at":"2026-07-05T04:32:30.242020+00:00","updated_at":"2026-07-05T04:32:30.242020+00:00"}