{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2019:7SO4UT3YLQBQ5QO7S34NUPBIK6","short_pith_number":"pith:7SO4UT3Y","schema_version":"1.0","canonical_sha256":"fc9dca4f785c030ec1df96f8da3c2857a433f3f004a2da6ca7ec2d1026855a73","source":{"kind":"arxiv","id":"1903.07120","version":5},"attestation_state":"computed","paper":{"title":"Stabilize Deep ResNet with A Sharp Scaling Factor $\\tau$","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["stat.ML"],"primary_cat":"cs.LG","authors_text":"Da Yu, Huishuai Zhang, Mingyang Yi, Tie-Yan Liu, Wei Chen","submitted_at":"2019-03-17T16:15:56Z","abstract_excerpt":"We study the stability and convergence of training deep ResNets with gradient descent. Specifically, we show that the parametric branch in the residual block should be scaled down by a factor $\\tau =O(1/\\sqrt{L})$ to guarantee stable forward/backward process, where $L$ is the number of residual blocks. Moreover, we establish a converse result that the forward process is unbounded when $\\tau>L^{-\\frac{1}{2}+c}$, for any positive constant $c$. The above two results together establish a sharp value of the scaling factor in determining the stability of deep ResNet. Based on the stability result, w"},"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":"1903.07120","kind":"arxiv","version":5},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2019-03-17T16:15:56Z","cross_cats_sorted":["stat.ML"],"title_canon_sha256":"3adc1fd43c5568d0c715d1ea10c33e0eb2e83072ec2c6fcbdcbb366825872be0","abstract_canon_sha256":"870dc40e48e80101ed80292d732bcbce3bf86e57e07d4f8453560701ef7f2514"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T05:37:02.324760Z","signature_b64":"kT1tNNPyXoq7rAGgLwTWohVSGez7iMp3sLTt6nn/p0UdziF19lSsCDyi44LIhZVOCGlkLho6zUN8xJjTUDgWCw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"fc9dca4f785c030ec1df96f8da3c2857a433f3f004a2da6ca7ec2d1026855a73","last_reissued_at":"2026-07-05T05:37:02.323625Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T05:37:02.323625Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Stabilize Deep ResNet with A Sharp Scaling Factor $\\tau$","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["stat.ML"],"primary_cat":"cs.LG","authors_text":"Da Yu, Huishuai Zhang, Mingyang Yi, Tie-Yan Liu, Wei Chen","submitted_at":"2019-03-17T16:15:56Z","abstract_excerpt":"We study the stability and convergence of training deep ResNets with gradient descent. Specifically, we show that the parametric branch in the residual block should be scaled down by a factor $\\tau =O(1/\\sqrt{L})$ to guarantee stable forward/backward process, where $L$ is the number of residual blocks. Moreover, we establish a converse result that the forward process is unbounded when $\\tau>L^{-\\frac{1}{2}+c}$, for any positive constant $c$. The above two results together establish a sharp value of the scaling factor in determining the stability of deep ResNet. Based on the stability result, w"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"1903.07120","kind":"arxiv","version":5},"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/1903.07120/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":"1903.07120","created_at":"2026-07-05T05:37:02.324253+00:00"},{"alias_kind":"arxiv_version","alias_value":"1903.07120v5","created_at":"2026-07-05T05:37:02.324253+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.1903.07120","created_at":"2026-07-05T05:37:02.324253+00:00"},{"alias_kind":"pith_short_12","alias_value":"7SO4UT3YLQBQ","created_at":"2026-07-05T05:37:02.324253+00:00"},{"alias_kind":"pith_short_16","alias_value":"7SO4UT3YLQBQ5QO7","created_at":"2026-07-05T05:37:02.324253+00:00"},{"alias_kind":"pith_short_8","alias_value":"7SO4UT3Y","created_at":"2026-07-05T05:37:02.324253+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":2,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.10089","citing_title":"A Theory on Flow Matching with Neural Networks","ref_index":258,"is_internal_anchor":false},{"citing_arxiv_id":"2401.01335","citing_title":"Self-Play Fine-Tuning Converts Weak Language Models to Strong Language Models","ref_index":239,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/7SO4UT3YLQBQ5QO7S34NUPBIK6","json":"https://pith.science/pith/7SO4UT3YLQBQ5QO7S34NUPBIK6.json","graph_json":"https://pith.science/api/pith-number/7SO4UT3YLQBQ5QO7S34NUPBIK6/graph.json","events_json":"https://pith.science/api/pith-number/7SO4UT3YLQBQ5QO7S34NUPBIK6/events.json","paper":"https://pith.science/paper/7SO4UT3Y"},"agent_actions":{"view_html":"https://pith.science/pith/7SO4UT3YLQBQ5QO7S34NUPBIK6","download_json":"https://pith.science/pith/7SO4UT3YLQBQ5QO7S34NUPBIK6.json","view_paper":"https://pith.science/paper/7SO4UT3Y","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=1903.07120&json=true","fetch_graph":"https://pith.science/api/pith-number/7SO4UT3YLQBQ5QO7S34NUPBIK6/graph.json","fetch_events":"https://pith.science/api/pith-number/7SO4UT3YLQBQ5QO7S34NUPBIK6/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/7SO4UT3YLQBQ5QO7S34NUPBIK6/action/timestamp_anchor","attest_storage":"https://pith.science/pith/7SO4UT3YLQBQ5QO7S34NUPBIK6/action/storage_attestation","attest_author":"https://pith.science/pith/7SO4UT3YLQBQ5QO7S34NUPBIK6/action/author_attestation","sign_citation":"https://pith.science/pith/7SO4UT3YLQBQ5QO7S34NUPBIK6/action/citation_signature","submit_replication":"https://pith.science/pith/7SO4UT3YLQBQ5QO7S34NUPBIK6/action/replication_record"}},"created_at":"2026-07-05T05:37:02.324253+00:00","updated_at":"2026-07-05T05:37:02.324253+00:00"}