{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:EE5TFPP2NEFF62DKAY2AKDLWFO","short_pith_number":"pith:EE5TFPP2","schema_version":"1.0","canonical_sha256":"213b32bdfa690a5f686a0634050d762bb53ec3b6c5e5a88955aabcfd03795877","source":{"kind":"arxiv","id":"2403.03942","version":2},"attestation_state":"computed","paper":{"title":"The Heuristic Core: Understanding Subnetwork Generalization in Pretrained Language Models","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.CL","authors_text":"Adithya Bhaskar, Dan Friedman, Danqi Chen","submitted_at":"2024-03-06T18:50:14Z","abstract_excerpt":"Prior work has found that pretrained language models (LMs) fine-tuned with different random seeds can achieve similar in-domain performance but generalize differently on tests of syntactic generalization. In this work, we show that, even within a single model, we can find multiple subnetworks that perform similarly in-domain, but generalize vastly differently. To better understand these phenomena, we investigate if they can be understood in terms of \"competing subnetworks\": the model initially represents a variety of distinct algorithms, corresponding to different subnetworks, and generalizati"},"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":"2403.03942","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2024-03-06T18:50:14Z","cross_cats_sorted":["cs.LG"],"title_canon_sha256":"aed8c4dbad411622052f834eeef5cce8d284bb7a0b01dfe926d6b1a2adc7bb47","abstract_canon_sha256":"ee4b2160feea116843ce435a80cba98d536dc3e30c2e2be718129c96de983e51"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T08:27:47.082229Z","signature_b64":"39rOMOqCvQyQo54UfAF1H++KlgPhwIUzFtfXApDLo7xMbPn5Zg96fYMVRNnSBrdc2pCpI6YLm4R+Kr79JBCHBA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"213b32bdfa690a5f686a0634050d762bb53ec3b6c5e5a88955aabcfd03795877","last_reissued_at":"2026-07-05T08:27:47.081712Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T08:27:47.081712Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"The Heuristic Core: Understanding Subnetwork Generalization in Pretrained Language Models","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.CL","authors_text":"Adithya Bhaskar, Dan Friedman, Danqi Chen","submitted_at":"2024-03-06T18:50:14Z","abstract_excerpt":"Prior work has found that pretrained language models (LMs) fine-tuned with different random seeds can achieve similar in-domain performance but generalize differently on tests of syntactic generalization. In this work, we show that, even within a single model, we can find multiple subnetworks that perform similarly in-domain, but generalize vastly differently. To better understand these phenomena, we investigate if they can be understood in terms of \"competing subnetworks\": the model initially represents a variety of distinct algorithms, corresponding to different subnetworks, and generalizati"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2403.03942","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/2403.03942/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":"2403.03942","created_at":"2026-07-05T08:27:47.081776+00:00"},{"alias_kind":"arxiv_version","alias_value":"2403.03942v2","created_at":"2026-07-05T08:27:47.081776+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2403.03942","created_at":"2026-07-05T08:27:47.081776+00:00"},{"alias_kind":"pith_short_12","alias_value":"EE5TFPP2NEFF","created_at":"2026-07-05T08:27:47.081776+00:00"},{"alias_kind":"pith_short_16","alias_value":"EE5TFPP2NEFF62DK","created_at":"2026-07-05T08:27:47.081776+00:00"},{"alias_kind":"pith_short_8","alias_value":"EE5TFPP2","created_at":"2026-07-05T08:27:47.081776+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2411.09475","citing_title":"ResidualDroppath: Enhancing Feature Reuse over Residual Connections","ref_index":9,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/EE5TFPP2NEFF62DKAY2AKDLWFO","json":"https://pith.science/pith/EE5TFPP2NEFF62DKAY2AKDLWFO.json","graph_json":"https://pith.science/api/pith-number/EE5TFPP2NEFF62DKAY2AKDLWFO/graph.json","events_json":"https://pith.science/api/pith-number/EE5TFPP2NEFF62DKAY2AKDLWFO/events.json","paper":"https://pith.science/paper/EE5TFPP2"},"agent_actions":{"view_html":"https://pith.science/pith/EE5TFPP2NEFF62DKAY2AKDLWFO","download_json":"https://pith.science/pith/EE5TFPP2NEFF62DKAY2AKDLWFO.json","view_paper":"https://pith.science/paper/EE5TFPP2","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2403.03942&json=true","fetch_graph":"https://pith.science/api/pith-number/EE5TFPP2NEFF62DKAY2AKDLWFO/graph.json","fetch_events":"https://pith.science/api/pith-number/EE5TFPP2NEFF62DKAY2AKDLWFO/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/EE5TFPP2NEFF62DKAY2AKDLWFO/action/timestamp_anchor","attest_storage":"https://pith.science/pith/EE5TFPP2NEFF62DKAY2AKDLWFO/action/storage_attestation","attest_author":"https://pith.science/pith/EE5TFPP2NEFF62DKAY2AKDLWFO/action/author_attestation","sign_citation":"https://pith.science/pith/EE5TFPP2NEFF62DKAY2AKDLWFO/action/citation_signature","submit_replication":"https://pith.science/pith/EE5TFPP2NEFF62DKAY2AKDLWFO/action/replication_record"}},"created_at":"2026-07-05T08:27:47.081776+00:00","updated_at":"2026-07-05T08:27:47.081776+00:00"}