{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2023:YW2LCWOF27DMFEBQY3X2VDJM3K","short_pith_number":"pith:YW2LCWOF","schema_version":"1.0","canonical_sha256":"c5b4b159c5d7c6c29030c6efaa8d2cda9963501e4ffadbce10e9e8963d6bfce1","source":{"kind":"arxiv","id":"2306.11668","version":1},"attestation_state":"computed","paper":{"title":"Principles for Initialization and Architecture Selection in Graph Neural Networks with ReLU Activations","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.LG","hep-ex","math.PR"],"primary_cat":"stat.ML","authors_text":"Boris Hanin, Gage DeZoort","submitted_at":"2023-06-20T16:40:41Z","abstract_excerpt":"This article derives and validates three principles for initialization and architecture selection in finite width graph neural networks (GNNs) with ReLU activations. First, we theoretically derive what is essentially the unique generalization to ReLU GNNs of the well-known He-initialization. Our initialization scheme guarantees that the average scale of network outputs and gradients remains order one at initialization. Second, we prove in finite width vanilla ReLU GNNs that oversmoothing is unavoidable at large depth when using fixed aggregation operator, regardless of initialization. We then "},"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":"2306.11668","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"stat.ML","submitted_at":"2023-06-20T16:40:41Z","cross_cats_sorted":["cs.AI","cs.LG","hep-ex","math.PR"],"title_canon_sha256":"821682fbedcd6a0b5441bb2166606e3fda4aee3356c77473d58edab91b6732f4","abstract_canon_sha256":"9d352d45f856bf146c18c1aa2397b17ce838ce7bb2ead7a7fba97a3355385117"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T06:22:22.339153Z","signature_b64":"4DTpgDpkSxlbIzuEANBfAbIFd/9b8pzh22rDTOOc0DbRvlWUAItfiW3oiImAFCphHfY0EomR0ZJyA9oNMpsKAw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"c5b4b159c5d7c6c29030c6efaa8d2cda9963501e4ffadbce10e9e8963d6bfce1","last_reissued_at":"2026-07-05T06:22:22.338703Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T06:22:22.338703Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Principles for Initialization and Architecture Selection in Graph Neural Networks with ReLU Activations","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.LG","hep-ex","math.PR"],"primary_cat":"stat.ML","authors_text":"Boris Hanin, Gage DeZoort","submitted_at":"2023-06-20T16:40:41Z","abstract_excerpt":"This article derives and validates three principles for initialization and architecture selection in finite width graph neural networks (GNNs) with ReLU activations. First, we theoretically derive what is essentially the unique generalization to ReLU GNNs of the well-known He-initialization. Our initialization scheme guarantees that the average scale of network outputs and gradients remains order one at initialization. Second, we prove in finite width vanilla ReLU GNNs that oversmoothing is unavoidable at large depth when using fixed aggregation operator, regardless of initialization. We then "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2306.11668","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/2306.11668/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":"2306.11668","created_at":"2026-07-05T06:22:22.338762+00:00"},{"alias_kind":"arxiv_version","alias_value":"2306.11668v1","created_at":"2026-07-05T06:22:22.338762+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2306.11668","created_at":"2026-07-05T06:22:22.338762+00:00"},{"alias_kind":"pith_short_12","alias_value":"YW2LCWOF27DM","created_at":"2026-07-05T06:22:22.338762+00:00"},{"alias_kind":"pith_short_16","alias_value":"YW2LCWOF27DMFEBQ","created_at":"2026-07-05T06:22:22.338762+00:00"},{"alias_kind":"pith_short_8","alias_value":"YW2LCWOF","created_at":"2026-07-05T06:22:22.338762+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2501.00762","citing_title":"Residual connections provably mitigate oversmoothing in graph neural networks","ref_index":6,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/YW2LCWOF27DMFEBQY3X2VDJM3K","json":"https://pith.science/pith/YW2LCWOF27DMFEBQY3X2VDJM3K.json","graph_json":"https://pith.science/api/pith-number/YW2LCWOF27DMFEBQY3X2VDJM3K/graph.json","events_json":"https://pith.science/api/pith-number/YW2LCWOF27DMFEBQY3X2VDJM3K/events.json","paper":"https://pith.science/paper/YW2LCWOF"},"agent_actions":{"view_html":"https://pith.science/pith/YW2LCWOF27DMFEBQY3X2VDJM3K","download_json":"https://pith.science/pith/YW2LCWOF27DMFEBQY3X2VDJM3K.json","view_paper":"https://pith.science/paper/YW2LCWOF","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2306.11668&json=true","fetch_graph":"https://pith.science/api/pith-number/YW2LCWOF27DMFEBQY3X2VDJM3K/graph.json","fetch_events":"https://pith.science/api/pith-number/YW2LCWOF27DMFEBQY3X2VDJM3K/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/YW2LCWOF27DMFEBQY3X2VDJM3K/action/timestamp_anchor","attest_storage":"https://pith.science/pith/YW2LCWOF27DMFEBQY3X2VDJM3K/action/storage_attestation","attest_author":"https://pith.science/pith/YW2LCWOF27DMFEBQY3X2VDJM3K/action/author_attestation","sign_citation":"https://pith.science/pith/YW2LCWOF27DMFEBQY3X2VDJM3K/action/citation_signature","submit_replication":"https://pith.science/pith/YW2LCWOF27DMFEBQY3X2VDJM3K/action/replication_record"}},"created_at":"2026-07-05T06:22:22.338762+00:00","updated_at":"2026-07-05T06:22:22.338762+00:00"}