{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:FXCGAPS6ZEMAE7QFM3Y2PM4U3D","short_pith_number":"pith:FXCGAPS6","schema_version":"1.0","canonical_sha256":"2dc4603e5ec918027e0566f1a7b394d8ddfb330b4bf6e16798ca473c833991ed","source":{"kind":"arxiv","id":"2501.08537","version":1},"attestation_state":"computed","paper":{"title":"Complexity Control Facilitates Reasoning-Based Compositional Generalization in Transformers","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.CL","authors_text":"Pengxiao Lin, Yaoyu Zhang, Zhi-Qin John Xu, Zhiwei Wang, Zhongwang Zhang","submitted_at":"2025-01-15T02:54:52Z","abstract_excerpt":"Transformers have demonstrated impressive capabilities across various tasks, yet their performance on compositional problems remains a subject of debate. In this study, we investigate the internal mechanisms underlying Transformers' behavior in compositional tasks. We find that complexity control strategies significantly influence whether the model learns primitive-level rules that generalize out-of-distribution (reasoning-based solutions) or relies solely on memorized mappings (memory-based solutions). By applying masking strategies to the model's information circuits and employing multiple c"},"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":"2501.08537","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2025-01-15T02:54:52Z","cross_cats_sorted":["cs.LG"],"title_canon_sha256":"13a054c395e3232986dd8eca5270efba329a492e6f6fd6c785edeb8cfe8d91d4","abstract_canon_sha256":"f13fe01fb267e8f6db566f58117a78ca22d5a8f78e5b01dfa2bc3594523703f1"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T10:01:20.866869Z","signature_b64":"MvGZEAZLaHV/usdpuV74auzMmNTrB2OOqs+qiIR3nBYIMcZsATfta0JKyMTkgv5Cbz18Xi+QpHagISpQh6bhDA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"2dc4603e5ec918027e0566f1a7b394d8ddfb330b4bf6e16798ca473c833991ed","last_reissued_at":"2026-07-05T10:01:20.866452Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T10:01:20.866452Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Complexity Control Facilitates Reasoning-Based Compositional Generalization in Transformers","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.CL","authors_text":"Pengxiao Lin, Yaoyu Zhang, Zhi-Qin John Xu, Zhiwei Wang, Zhongwang Zhang","submitted_at":"2025-01-15T02:54:52Z","abstract_excerpt":"Transformers have demonstrated impressive capabilities across various tasks, yet their performance on compositional problems remains a subject of debate. In this study, we investigate the internal mechanisms underlying Transformers' behavior in compositional tasks. We find that complexity control strategies significantly influence whether the model learns primitive-level rules that generalize out-of-distribution (reasoning-based solutions) or relies solely on memorized mappings (memory-based solutions). By applying masking strategies to the model's information circuits and employing multiple c"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2501.08537","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/2501.08537/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":"2501.08537","created_at":"2026-07-05T10:01:20.866513+00:00"},{"alias_kind":"arxiv_version","alias_value":"2501.08537v1","created_at":"2026-07-05T10:01:20.866513+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2501.08537","created_at":"2026-07-05T10:01:20.866513+00:00"},{"alias_kind":"pith_short_12","alias_value":"FXCGAPS6ZEMA","created_at":"2026-07-05T10:01:20.866513+00:00"},{"alias_kind":"pith_short_16","alias_value":"FXCGAPS6ZEMAE7QF","created_at":"2026-07-05T10:01:20.866513+00:00"},{"alias_kind":"pith_short_8","alias_value":"FXCGAPS6","created_at":"2026-07-05T10:01:20.866513+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":2,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2607.00452","citing_title":"Gauging, Measuring, and Controlling Critic Complexity in Actor-Critic Reinforcement Learning","ref_index":11,"is_internal_anchor":false},{"citing_arxiv_id":"2504.09484","citing_title":"An overview of condensation phenomenon in deep learning","ref_index":18,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/FXCGAPS6ZEMAE7QFM3Y2PM4U3D","json":"https://pith.science/pith/FXCGAPS6ZEMAE7QFM3Y2PM4U3D.json","graph_json":"https://pith.science/api/pith-number/FXCGAPS6ZEMAE7QFM3Y2PM4U3D/graph.json","events_json":"https://pith.science/api/pith-number/FXCGAPS6ZEMAE7QFM3Y2PM4U3D/events.json","paper":"https://pith.science/paper/FXCGAPS6"},"agent_actions":{"view_html":"https://pith.science/pith/FXCGAPS6ZEMAE7QFM3Y2PM4U3D","download_json":"https://pith.science/pith/FXCGAPS6ZEMAE7QFM3Y2PM4U3D.json","view_paper":"https://pith.science/paper/FXCGAPS6","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2501.08537&json=true","fetch_graph":"https://pith.science/api/pith-number/FXCGAPS6ZEMAE7QFM3Y2PM4U3D/graph.json","fetch_events":"https://pith.science/api/pith-number/FXCGAPS6ZEMAE7QFM3Y2PM4U3D/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/FXCGAPS6ZEMAE7QFM3Y2PM4U3D/action/timestamp_anchor","attest_storage":"https://pith.science/pith/FXCGAPS6ZEMAE7QFM3Y2PM4U3D/action/storage_attestation","attest_author":"https://pith.science/pith/FXCGAPS6ZEMAE7QFM3Y2PM4U3D/action/author_attestation","sign_citation":"https://pith.science/pith/FXCGAPS6ZEMAE7QFM3Y2PM4U3D/action/citation_signature","submit_replication":"https://pith.science/pith/FXCGAPS6ZEMAE7QFM3Y2PM4U3D/action/replication_record"}},"created_at":"2026-07-05T10:01:20.866513+00:00","updated_at":"2026-07-05T10:01:20.866513+00:00"}