{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:Y5K7QRPFDUB6XPCIV7T7TAKIW5","short_pith_number":"pith:Y5K7QRPF","schema_version":"1.0","canonical_sha256":"c755f845e51d03ebbc48afe7f98148b77190fd4ccb1a5bc571230523f84e7240","source":{"kind":"arxiv","id":"2405.05409","version":5},"attestation_state":"computed","paper":{"title":"Initialization is Critical to Whether Transformers Fit Composite Functions by Reasoning or Memorizing","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Pengxiao Lin, Yaoyu Zhang, Zhi-Qin John Xu, Zhiwei Wang, Zhongwang Zhang","submitted_at":"2024-05-08T20:23:24Z","abstract_excerpt":"Transformers have shown impressive capabilities across various tasks, but their performance on compositional problems remains a topic of debate. In this work, we investigate the mechanisms of how transformers behave on unseen compositional tasks. We discover that the parameter initialization scale plays a critical role in determining whether the model learns inferential (reasoning-based) solutions, which capture the underlying compositional primitives, or symmetric (memory-based) solutions, which simply memorize mappings without understanding the compositional structure. By analyzing the infor"},"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":"2405.05409","kind":"arxiv","version":5},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2024-05-08T20:23:24Z","cross_cats_sorted":[],"title_canon_sha256":"57a549d2586fbfebca24ede5f5aa43f924237047d08a5ac5c2aea641eab7c2d6","abstract_canon_sha256":"c1a0b7a1e9bdb1c7ad2091c65d879820a665aaf2cb9bb7d6d8f97f0b36a5e4b8"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T09:59:46.520969Z","signature_b64":"9SK+to12yx9nLsIedwIDqyLyCI5EqYaqOASByYY4Jd5TtLqpWy9sFMYRNjZAKmkknHAhqKgHCV9ooRgkmNOtDQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"c755f845e51d03ebbc48afe7f98148b77190fd4ccb1a5bc571230523f84e7240","last_reissued_at":"2026-07-05T09:59:46.520588Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T09:59:46.520588Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Initialization is Critical to Whether Transformers Fit Composite Functions by Reasoning or Memorizing","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Pengxiao Lin, Yaoyu Zhang, Zhi-Qin John Xu, Zhiwei Wang, Zhongwang Zhang","submitted_at":"2024-05-08T20:23:24Z","abstract_excerpt":"Transformers have shown impressive capabilities across various tasks, but their performance on compositional problems remains a topic of debate. In this work, we investigate the mechanisms of how transformers behave on unseen compositional tasks. We discover that the parameter initialization scale plays a critical role in determining whether the model learns inferential (reasoning-based) solutions, which capture the underlying compositional primitives, or symmetric (memory-based) solutions, which simply memorize mappings without understanding the compositional structure. By analyzing the infor"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2405.05409","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/2405.05409/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":"2405.05409","created_at":"2026-07-05T09:59:46.520649+00:00"},{"alias_kind":"arxiv_version","alias_value":"2405.05409v5","created_at":"2026-07-05T09:59:46.520649+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2405.05409","created_at":"2026-07-05T09:59:46.520649+00:00"},{"alias_kind":"pith_short_12","alias_value":"Y5K7QRPFDUB6","created_at":"2026-07-05T09:59:46.520649+00:00"},{"alias_kind":"pith_short_16","alias_value":"Y5K7QRPFDUB6XPCI","created_at":"2026-07-05T09:59:46.520649+00:00"},{"alias_kind":"pith_short_8","alias_value":"Y5K7QRPF","created_at":"2026-07-05T09:59:46.520649+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2605.03667","citing_title":"ELAS: Efficient Pre-Training of Low-Rank Large Language Models via 2:4 Activation Sparsity","ref_index":14,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/Y5K7QRPFDUB6XPCIV7T7TAKIW5","json":"https://pith.science/pith/Y5K7QRPFDUB6XPCIV7T7TAKIW5.json","graph_json":"https://pith.science/api/pith-number/Y5K7QRPFDUB6XPCIV7T7TAKIW5/graph.json","events_json":"https://pith.science/api/pith-number/Y5K7QRPFDUB6XPCIV7T7TAKIW5/events.json","paper":"https://pith.science/paper/Y5K7QRPF"},"agent_actions":{"view_html":"https://pith.science/pith/Y5K7QRPFDUB6XPCIV7T7TAKIW5","download_json":"https://pith.science/pith/Y5K7QRPFDUB6XPCIV7T7TAKIW5.json","view_paper":"https://pith.science/paper/Y5K7QRPF","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2405.05409&json=true","fetch_graph":"https://pith.science/api/pith-number/Y5K7QRPFDUB6XPCIV7T7TAKIW5/graph.json","fetch_events":"https://pith.science/api/pith-number/Y5K7QRPFDUB6XPCIV7T7TAKIW5/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/Y5K7QRPFDUB6XPCIV7T7TAKIW5/action/timestamp_anchor","attest_storage":"https://pith.science/pith/Y5K7QRPFDUB6XPCIV7T7TAKIW5/action/storage_attestation","attest_author":"https://pith.science/pith/Y5K7QRPFDUB6XPCIV7T7TAKIW5/action/author_attestation","sign_citation":"https://pith.science/pith/Y5K7QRPFDUB6XPCIV7T7TAKIW5/action/citation_signature","submit_replication":"https://pith.science/pith/Y5K7QRPFDUB6XPCIV7T7TAKIW5/action/replication_record"}},"created_at":"2026-07-05T09:59:46.520649+00:00","updated_at":"2026-07-05T09:59:46.520649+00:00"}