{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2018:4NARQFBC6XNKSWAJAILEEWYW37","short_pith_number":"pith:4NARQFBC","schema_version":"1.0","canonical_sha256":"e341181422f5daa958090216425b16dfe313a4faa5b73c5438cc1bfa3b340973","source":{"kind":"arxiv","id":"1803.01719","version":3},"attestation_state":"computed","paper":{"title":"How to Start Training: The Effect of Initialization and Architecture","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"stat.ML","authors_text":"Boris Hanin, David Rolnick","submitted_at":"2018-03-05T15:17:50Z","abstract_excerpt":"We identify and study two common failure modes for early training in deep ReLU nets. For each we give a rigorous proof of when it occurs and how to avoid it, for fully connected and residual architectures. The first failure mode, exploding/vanishing mean activation length, can be avoided by initializing weights from a symmetric distribution with variance 2/fan-in and, for ResNets, by correctly weighting the residual modules. We prove that the second failure mode, exponentially large variance of activation length, never occurs in residual nets once the first failure mode is avoided. In contrast"},"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":"1803.01719","kind":"arxiv","version":3},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"stat.ML","submitted_at":"2018-03-05T15:17:50Z","cross_cats_sorted":["cs.LG"],"title_canon_sha256":"5b1b8a655abcfa2d4c5e2fbe4a569933e259bd42b093b0f3134828684cd06bcd","abstract_canon_sha256":"f68d02c781947a28719f058dafe60b010656bc375453a4ff075e3e3fd4c01d7c"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-18T00:00:57.164418Z","signature_b64":"eXiYYEjbiCWI/CmDOnOEW23JEFUFdd/4mY+f5vXzZx9bQccZWJWfisNK29jTvVPiHTUmxGco9nxj95cWOaCDBg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"e341181422f5daa958090216425b16dfe313a4faa5b73c5438cc1bfa3b340973","last_reissued_at":"2026-05-18T00:00:57.163791Z","signature_status":"signed_v1","first_computed_at":"2026-05-18T00:00:57.163791Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"How to Start Training: The Effect of Initialization and Architecture","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"stat.ML","authors_text":"Boris Hanin, David Rolnick","submitted_at":"2018-03-05T15:17:50Z","abstract_excerpt":"We identify and study two common failure modes for early training in deep ReLU nets. For each we give a rigorous proof of when it occurs and how to avoid it, for fully connected and residual architectures. The first failure mode, exploding/vanishing mean activation length, can be avoided by initializing weights from a symmetric distribution with variance 2/fan-in and, for ResNets, by correctly weighting the residual modules. We prove that the second failure mode, exponentially large variance of activation length, never occurs in residual nets once the first failure mode is avoided. In contrast"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"1803.01719","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":""},"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":"1803.01719","created_at":"2026-05-18T00:00:57.163883+00:00"},{"alias_kind":"arxiv_version","alias_value":"1803.01719v3","created_at":"2026-05-18T00:00:57.163883+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.1803.01719","created_at":"2026-05-18T00:00:57.163883+00:00"},{"alias_kind":"pith_short_12","alias_value":"4NARQFBC6XNK","created_at":"2026-05-18T12:32:05.422762+00:00"},{"alias_kind":"pith_short_16","alias_value":"4NARQFBC6XNKSWAJ","created_at":"2026-05-18T12:32:05.422762+00:00"},{"alias_kind":"pith_short_8","alias_value":"4NARQFBC","created_at":"2026-05-18T12:32:05.422762+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2604.09391","citing_title":"Efficient Unlearning through Maximizing Relearning Convergence Delay","ref_index":23,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/4NARQFBC6XNKSWAJAILEEWYW37","json":"https://pith.science/pith/4NARQFBC6XNKSWAJAILEEWYW37.json","graph_json":"https://pith.science/api/pith-number/4NARQFBC6XNKSWAJAILEEWYW37/graph.json","events_json":"https://pith.science/api/pith-number/4NARQFBC6XNKSWAJAILEEWYW37/events.json","paper":"https://pith.science/paper/4NARQFBC"},"agent_actions":{"view_html":"https://pith.science/pith/4NARQFBC6XNKSWAJAILEEWYW37","download_json":"https://pith.science/pith/4NARQFBC6XNKSWAJAILEEWYW37.json","view_paper":"https://pith.science/paper/4NARQFBC","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=1803.01719&json=true","fetch_graph":"https://pith.science/api/pith-number/4NARQFBC6XNKSWAJAILEEWYW37/graph.json","fetch_events":"https://pith.science/api/pith-number/4NARQFBC6XNKSWAJAILEEWYW37/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/4NARQFBC6XNKSWAJAILEEWYW37/action/timestamp_anchor","attest_storage":"https://pith.science/pith/4NARQFBC6XNKSWAJAILEEWYW37/action/storage_attestation","attest_author":"https://pith.science/pith/4NARQFBC6XNKSWAJAILEEWYW37/action/author_attestation","sign_citation":"https://pith.science/pith/4NARQFBC6XNKSWAJAILEEWYW37/action/citation_signature","submit_replication":"https://pith.science/pith/4NARQFBC6XNKSWAJAILEEWYW37/action/replication_record"}},"created_at":"2026-05-18T00:00:57.163883+00:00","updated_at":"2026-05-18T00:00:57.163883+00:00"}