{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:NLVJQULRUIZMJEMZ4LYDVRHKUU","short_pith_number":"pith:NLVJQULR","schema_version":"1.0","canonical_sha256":"6aea985171a232c49199e2f03ac4eaa50302ee09a8ed37413feb585ddaed21ec","source":{"kind":"arxiv","id":"2605.27946","version":1},"attestation_state":"computed","paper":{"title":"Is Backpropagation Optimal? When Synthetic Gradients Improve Sample Efficiency","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"stat.ML","authors_text":"Sanmi Koyejo, Yibo Jacky Zhang, Zeyu Tang","submitted_at":"2026-05-27T04:37:04Z","abstract_excerpt":"Backpropagation is the default learning rule for artificial neural networks and is often treated as the settled approach whenever differentiability is available. In this work, we revisit this convention through a theoretical lens of sample efficiency. We introduce a unified vectorized feedback framework for loss-based and reward-based learning on computational graphs, in which synthetic gradients emerge as a natural alternative to backpropagation. We characterize the conditions under which synthetic gradients can achieve a lower gradient-estimation mean squared error than backpropagation. We c"},"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":"2605.27946","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"stat.ML","submitted_at":"2026-05-27T04:37:04Z","cross_cats_sorted":["cs.LG"],"title_canon_sha256":"acdf3c1e62ec88aae13237c7e9236640ad0c0ce22fa71b8a4bbba4f7b4b72093","abstract_canon_sha256":"7bf41dec24519f4b17f50574e8cf84be55d2e562314eb806fbe0277adc4e0873"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-28T01:04:53.594761Z","signature_b64":"c9pwRbOSq5ELpr/2qpjnkRZQYRSjdu/Gzsv1M7rD0DhzGjlECZuiOKhif+iw+9vYawiEdIVlM09wlOJv6zkLBQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"6aea985171a232c49199e2f03ac4eaa50302ee09a8ed37413feb585ddaed21ec","last_reissued_at":"2026-05-28T01:04:53.594320Z","signature_status":"signed_v1","first_computed_at":"2026-05-28T01:04:53.594320Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Is Backpropagation Optimal? When Synthetic Gradients Improve Sample Efficiency","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"stat.ML","authors_text":"Sanmi Koyejo, Yibo Jacky Zhang, Zeyu Tang","submitted_at":"2026-05-27T04:37:04Z","abstract_excerpt":"Backpropagation is the default learning rule for artificial neural networks and is often treated as the settled approach whenever differentiability is available. In this work, we revisit this convention through a theoretical lens of sample efficiency. We introduce a unified vectorized feedback framework for loss-based and reward-based learning on computational graphs, in which synthetic gradients emerge as a natural alternative to backpropagation. We characterize the conditions under which synthetic gradients can achieve a lower gradient-estimation mean squared error than backpropagation. We c"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2605.27946","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/2605.27946/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":"2605.27946","created_at":"2026-05-28T01:04:53.594379+00:00"},{"alias_kind":"arxiv_version","alias_value":"2605.27946v1","created_at":"2026-05-28T01:04:53.594379+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.27946","created_at":"2026-05-28T01:04:53.594379+00:00"},{"alias_kind":"pith_short_12","alias_value":"NLVJQULRUIZM","created_at":"2026-05-28T01:04:53.594379+00:00"},{"alias_kind":"pith_short_16","alias_value":"NLVJQULRUIZMJEMZ","created_at":"2026-05-28T01:04:53.594379+00:00"},{"alias_kind":"pith_short_8","alias_value":"NLVJQULR","created_at":"2026-05-28T01:04:53.594379+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/NLVJQULRUIZMJEMZ4LYDVRHKUU","json":"https://pith.science/pith/NLVJQULRUIZMJEMZ4LYDVRHKUU.json","graph_json":"https://pith.science/api/pith-number/NLVJQULRUIZMJEMZ4LYDVRHKUU/graph.json","events_json":"https://pith.science/api/pith-number/NLVJQULRUIZMJEMZ4LYDVRHKUU/events.json","paper":"https://pith.science/paper/NLVJQULR"},"agent_actions":{"view_html":"https://pith.science/pith/NLVJQULRUIZMJEMZ4LYDVRHKUU","download_json":"https://pith.science/pith/NLVJQULRUIZMJEMZ4LYDVRHKUU.json","view_paper":"https://pith.science/paper/NLVJQULR","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2605.27946&json=true","fetch_graph":"https://pith.science/api/pith-number/NLVJQULRUIZMJEMZ4LYDVRHKUU/graph.json","fetch_events":"https://pith.science/api/pith-number/NLVJQULRUIZMJEMZ4LYDVRHKUU/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/NLVJQULRUIZMJEMZ4LYDVRHKUU/action/timestamp_anchor","attest_storage":"https://pith.science/pith/NLVJQULRUIZMJEMZ4LYDVRHKUU/action/storage_attestation","attest_author":"https://pith.science/pith/NLVJQULRUIZMJEMZ4LYDVRHKUU/action/author_attestation","sign_citation":"https://pith.science/pith/NLVJQULRUIZMJEMZ4LYDVRHKUU/action/citation_signature","submit_replication":"https://pith.science/pith/NLVJQULRUIZMJEMZ4LYDVRHKUU/action/replication_record"}},"created_at":"2026-05-28T01:04:53.594379+00:00","updated_at":"2026-05-28T01:04:53.594379+00:00"}