{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2019:D74W6D45UFR46GGWHZM4B2L53I","short_pith_number":"pith:D74W6D45","schema_version":"1.0","canonical_sha256":"1ff96f0f9da163cf18d63e59c0e97dda132419aac404c471725ee0ba5db9fba4","source":{"kind":"arxiv","id":"1906.09108","version":3},"attestation_state":"computed","paper":{"title":"Fully Decoupled Neural Network Learning Using Delayed Gradients","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CV","authors_text":"Huiping Zhuang, Qinglai Liu, Shuai Zhang, Yi Wang, Zhiping Lin","submitted_at":"2019-06-21T13:02:35Z","abstract_excerpt":"Training neural networks with back-propagation (BP) requires a sequential passing of activations and gradients, which forces the network modules to work in a synchronous fashion. This has been recognized as the lockings (i.e., the forward, backward and update lockings) inherited from the BP. In this paper, we propose a fully decoupled training scheme using delayed gradients (FDG) to break all these lockings. The FDG splits a neural network into multiple modules and trains them independently and asynchronously using different workers (e.g., GPUs). We also introduce a gradient shrinking process "},"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":"1906.09108","kind":"arxiv","version":3},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CV","submitted_at":"2019-06-21T13:02:35Z","cross_cats_sorted":[],"title_canon_sha256":"d9065a367959ef763de49427361332f790694756a426589e30e60aff286a3f56","abstract_canon_sha256":"3046875b641cba924ab9095cf349a2db4fc0f0e0829b3694c2ae2bbf7cf8a208"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T00:21:07.066703Z","signature_b64":"dVPMXtleAHGzLMeZiHJm9SZ8/FUYrAxZ59P/KP7MSypzu8oJDvYNCNM0HnS10Tp3ca0JkVQsdL43ITBEUvapDw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"1ff96f0f9da163cf18d63e59c0e97dda132419aac404c471725ee0ba5db9fba4","last_reissued_at":"2026-07-05T00:21:07.066259Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T00:21:07.066259Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Fully Decoupled Neural Network Learning Using Delayed Gradients","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CV","authors_text":"Huiping Zhuang, Qinglai Liu, Shuai Zhang, Yi Wang, Zhiping Lin","submitted_at":"2019-06-21T13:02:35Z","abstract_excerpt":"Training neural networks with back-propagation (BP) requires a sequential passing of activations and gradients, which forces the network modules to work in a synchronous fashion. This has been recognized as the lockings (i.e., the forward, backward and update lockings) inherited from the BP. In this paper, we propose a fully decoupled training scheme using delayed gradients (FDG) to break all these lockings. The FDG splits a neural network into multiple modules and trains them independently and asynchronously using different workers (e.g., GPUs). We also introduce a gradient shrinking process "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"1906.09108","kind":"arxiv","version":3},"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/1906.09108/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":"1906.09108","created_at":"2026-07-05T00:21:07.066311+00:00"},{"alias_kind":"arxiv_version","alias_value":"1906.09108v3","created_at":"2026-07-05T00:21:07.066311+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.1906.09108","created_at":"2026-07-05T00:21:07.066311+00:00"},{"alias_kind":"pith_short_12","alias_value":"D74W6D45UFR4","created_at":"2026-07-05T00:21:07.066311+00:00"},{"alias_kind":"pith_short_16","alias_value":"D74W6D45UFR46GGW","created_at":"2026-07-05T00:21:07.066311+00:00"},{"alias_kind":"pith_short_8","alias_value":"D74W6D45","created_at":"2026-07-05T00:21:07.066311+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.01787","citing_title":"Stochastic convergence of parallel asynchronous adaptive first-order methods","ref_index":55,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/D74W6D45UFR46GGWHZM4B2L53I","json":"https://pith.science/pith/D74W6D45UFR46GGWHZM4B2L53I.json","graph_json":"https://pith.science/api/pith-number/D74W6D45UFR46GGWHZM4B2L53I/graph.json","events_json":"https://pith.science/api/pith-number/D74W6D45UFR46GGWHZM4B2L53I/events.json","paper":"https://pith.science/paper/D74W6D45"},"agent_actions":{"view_html":"https://pith.science/pith/D74W6D45UFR46GGWHZM4B2L53I","download_json":"https://pith.science/pith/D74W6D45UFR46GGWHZM4B2L53I.json","view_paper":"https://pith.science/paper/D74W6D45","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=1906.09108&json=true","fetch_graph":"https://pith.science/api/pith-number/D74W6D45UFR46GGWHZM4B2L53I/graph.json","fetch_events":"https://pith.science/api/pith-number/D74W6D45UFR46GGWHZM4B2L53I/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/D74W6D45UFR46GGWHZM4B2L53I/action/timestamp_anchor","attest_storage":"https://pith.science/pith/D74W6D45UFR46GGWHZM4B2L53I/action/storage_attestation","attest_author":"https://pith.science/pith/D74W6D45UFR46GGWHZM4B2L53I/action/author_attestation","sign_citation":"https://pith.science/pith/D74W6D45UFR46GGWHZM4B2L53I/action/citation_signature","submit_replication":"https://pith.science/pith/D74W6D45UFR46GGWHZM4B2L53I/action/replication_record"}},"created_at":"2026-07-05T00:21:07.066311+00:00","updated_at":"2026-07-05T00:21:07.066311+00:00"}