{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2020:NWLBB6UGRU5DJAY2DILSVBAGXL","short_pith_number":"pith:NWLBB6UG","schema_version":"1.0","canonical_sha256":"6d9610fa868d3a34831a1a172a8406bad0749e5ddb1040ea422c5438b346d185","source":{"kind":"arxiv","id":"2009.08576","version":2},"attestation_state":"computed","paper":{"title":"Pruning Neural Networks at Initialization: Why are We Missing the Mark?","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.CV","cs.NE","stat.ML"],"primary_cat":"cs.LG","authors_text":"Daniel M. Roy, Gintare Karolina Dziugaite, Jonathan Frankle, Michael Carbin","submitted_at":"2020-09-18T01:13:38Z","abstract_excerpt":"Recent work has explored the possibility of pruning neural networks at initialization. We assess proposals for doing so: SNIP (Lee et al., 2019), GraSP (Wang et al., 2020), SynFlow (Tanaka et al., 2020), and magnitude pruning. Although these methods surpass the trivial baseline of random pruning, they remain below the accuracy of magnitude pruning after training, and we endeavor to understand why. We show that, unlike pruning after training, randomly shuffling the weights these methods prune within each layer or sampling new initial values preserves or improves accuracy. As such, the per-weigh"},"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":"2009.08576","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2020-09-18T01:13:38Z","cross_cats_sorted":["cs.CV","cs.NE","stat.ML"],"title_canon_sha256":"8bf665fe3d2c4eea9ba552e5f7d5bb9b102e406b65987f2a38759b639d83c392","abstract_canon_sha256":"b3183d8318c9847acd9dfa29841a42089479153d6ef174da8f838cf21357b7d4"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T02:24:51.309670Z","signature_b64":"gdWTZDV/W/IhiNS+SSG2GcZ10pMDrFH3BTA0wzT8vu4xZNhCGJMeiw7CzBguRlhNzSxixa5zk+yx/rqzM3opBg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"6d9610fa868d3a34831a1a172a8406bad0749e5ddb1040ea422c5438b346d185","last_reissued_at":"2026-07-05T02:24:51.309240Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T02:24:51.309240Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Pruning Neural Networks at Initialization: Why are We Missing the Mark?","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.CV","cs.NE","stat.ML"],"primary_cat":"cs.LG","authors_text":"Daniel M. Roy, Gintare Karolina Dziugaite, Jonathan Frankle, Michael Carbin","submitted_at":"2020-09-18T01:13:38Z","abstract_excerpt":"Recent work has explored the possibility of pruning neural networks at initialization. We assess proposals for doing so: SNIP (Lee et al., 2019), GraSP (Wang et al., 2020), SynFlow (Tanaka et al., 2020), and magnitude pruning. Although these methods surpass the trivial baseline of random pruning, they remain below the accuracy of magnitude pruning after training, and we endeavor to understand why. We show that, unlike pruning after training, randomly shuffling the weights these methods prune within each layer or sampling new initial values preserves or improves accuracy. As such, the per-weigh"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2009.08576","kind":"arxiv","version":2},"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/2009.08576/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":"2009.08576","created_at":"2026-07-05T02:24:51.309296+00:00"},{"alias_kind":"arxiv_version","alias_value":"2009.08576v2","created_at":"2026-07-05T02:24:51.309296+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2009.08576","created_at":"2026-07-05T02:24:51.309296+00:00"},{"alias_kind":"pith_short_12","alias_value":"NWLBB6UGRU5D","created_at":"2026-07-05T02:24:51.309296+00:00"},{"alias_kind":"pith_short_16","alias_value":"NWLBB6UGRU5DJAY2","created_at":"2026-07-05T02:24:51.309296+00:00"},{"alias_kind":"pith_short_8","alias_value":"NWLBB6UG","created_at":"2026-07-05T02:24:51.309296+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":2,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.22019","citing_title":"Channel Location Constrains the Auditability of Subliminal Learning","ref_index":25,"is_internal_anchor":false},{"citing_arxiv_id":"2310.12508","citing_title":"SalUn: Empowering Machine Unlearning via Gradient-based Weight Saliency in Both Image Classification and Generation","ref_index":166,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/NWLBB6UGRU5DJAY2DILSVBAGXL","json":"https://pith.science/pith/NWLBB6UGRU5DJAY2DILSVBAGXL.json","graph_json":"https://pith.science/api/pith-number/NWLBB6UGRU5DJAY2DILSVBAGXL/graph.json","events_json":"https://pith.science/api/pith-number/NWLBB6UGRU5DJAY2DILSVBAGXL/events.json","paper":"https://pith.science/paper/NWLBB6UG"},"agent_actions":{"view_html":"https://pith.science/pith/NWLBB6UGRU5DJAY2DILSVBAGXL","download_json":"https://pith.science/pith/NWLBB6UGRU5DJAY2DILSVBAGXL.json","view_paper":"https://pith.science/paper/NWLBB6UG","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2009.08576&json=true","fetch_graph":"https://pith.science/api/pith-number/NWLBB6UGRU5DJAY2DILSVBAGXL/graph.json","fetch_events":"https://pith.science/api/pith-number/NWLBB6UGRU5DJAY2DILSVBAGXL/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/NWLBB6UGRU5DJAY2DILSVBAGXL/action/timestamp_anchor","attest_storage":"https://pith.science/pith/NWLBB6UGRU5DJAY2DILSVBAGXL/action/storage_attestation","attest_author":"https://pith.science/pith/NWLBB6UGRU5DJAY2DILSVBAGXL/action/author_attestation","sign_citation":"https://pith.science/pith/NWLBB6UGRU5DJAY2DILSVBAGXL/action/citation_signature","submit_replication":"https://pith.science/pith/NWLBB6UGRU5DJAY2DILSVBAGXL/action/replication_record"}},"created_at":"2026-07-05T02:24:51.309296+00:00","updated_at":"2026-07-05T02:24:51.309296+00:00"}