{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:EKHB24BRLHG4CC5QH2Q4RWFV75","short_pith_number":"pith:EKHB24BR","schema_version":"1.0","canonical_sha256":"228e1d703159cdc10bb03ea1c8d8b5ff60a096b533fd616b4b765ebe86a47add","source":{"kind":"arxiv","id":"2505.23683","version":1},"attestation_state":"computed","paper":{"title":"Learning Compositional Functions with Transformers from Easy-to-Hard Data","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Alberto Bietti, Alex Damian, Daniel Hsu, Denny Wu, Eshaan Nichani, Jason D. Lee, Zixuan Wang","submitted_at":"2025-05-29T17:22:00Z","abstract_excerpt":"Transformer-based language models have demonstrated impressive capabilities across a range of complex reasoning tasks. Prior theoretical work exploring the expressive power of transformers has shown that they can efficiently perform multi-step reasoning tasks involving parallelizable computations. However, the learnability of such constructions, particularly the conditions on the data distribution that enable efficient learning via gradient-based optimization, remains an open question. Towards answering this question, in this work we study the learnability of the $k$-fold composition task, whi"},"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":"2505.23683","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2025-05-29T17:22:00Z","cross_cats_sorted":[],"title_canon_sha256":"620625165077d758f48a4a305c6d1df918757108d70e1414a26b6ef1c6d3ecf0","abstract_canon_sha256":"b41f515641e59658850b62a7f503072d24699961d9598cb5c0b01453d8ffb436"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:12:12.700869Z","signature_b64":"18iwECCNGOnkP7k8okkYw8m7aU3A0bpp3awcWXdyuo9N//y3hUSoCiy18nck6WWRP/bdxMcYa4+F/b0SLDNWDA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"228e1d703159cdc10bb03ea1c8d8b5ff60a096b533fd616b4b765ebe86a47add","last_reissued_at":"2026-07-05T11:12:12.700344Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:12:12.700344Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Learning Compositional Functions with Transformers from Easy-to-Hard Data","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Alberto Bietti, Alex Damian, Daniel Hsu, Denny Wu, Eshaan Nichani, Jason D. Lee, Zixuan Wang","submitted_at":"2025-05-29T17:22:00Z","abstract_excerpt":"Transformer-based language models have demonstrated impressive capabilities across a range of complex reasoning tasks. Prior theoretical work exploring the expressive power of transformers has shown that they can efficiently perform multi-step reasoning tasks involving parallelizable computations. However, the learnability of such constructions, particularly the conditions on the data distribution that enable efficient learning via gradient-based optimization, remains an open question. Towards answering this question, in this work we study the learnability of the $k$-fold composition task, whi"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2505.23683","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/2505.23683/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":"2505.23683","created_at":"2026-07-05T11:12:12.700408+00:00"},{"alias_kind":"arxiv_version","alias_value":"2505.23683v1","created_at":"2026-07-05T11:12:12.700408+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2505.23683","created_at":"2026-07-05T11:12:12.700408+00:00"},{"alias_kind":"pith_short_12","alias_value":"EKHB24BRLHG4","created_at":"2026-07-05T11:12:12.700408+00:00"},{"alias_kind":"pith_short_16","alias_value":"EKHB24BRLHG4CC5Q","created_at":"2026-07-05T11:12:12.700408+00:00"},{"alias_kind":"pith_short_8","alias_value":"EKHB24BR","created_at":"2026-07-05T11:12:12.700408+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":6,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2605.28600","citing_title":"Transformers Provably Learn to Internalize Chain-of-Thought","ref_index":47,"is_internal_anchor":false},{"citing_arxiv_id":"2510.26745","citing_title":"Deep sequence models tend to memorize geometrically; it is unclear why","ref_index":186,"is_internal_anchor":false},{"citing_arxiv_id":"2510.25741","citing_title":"Scaling Latent Reasoning via Looped Language Models","ref_index":81,"is_internal_anchor":false},{"citing_arxiv_id":"2603.26554","citing_title":"Sharp Capacity Scaling of Spectral Optimizers in Learning Associative Memory","ref_index":58,"is_internal_anchor":false},{"citing_arxiv_id":"2604.22951","citing_title":"The Power of Power Law: Asymmetry Enables Compositional Reasoning","ref_index":53,"is_internal_anchor":false},{"citing_arxiv_id":"2605.06609","citing_title":"Transformers Efficiently Perform In-Context Logistic Regression via Normalized Gradient Descent","ref_index":50,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/EKHB24BRLHG4CC5QH2Q4RWFV75","json":"https://pith.science/pith/EKHB24BRLHG4CC5QH2Q4RWFV75.json","graph_json":"https://pith.science/api/pith-number/EKHB24BRLHG4CC5QH2Q4RWFV75/graph.json","events_json":"https://pith.science/api/pith-number/EKHB24BRLHG4CC5QH2Q4RWFV75/events.json","paper":"https://pith.science/paper/EKHB24BR"},"agent_actions":{"view_html":"https://pith.science/pith/EKHB24BRLHG4CC5QH2Q4RWFV75","download_json":"https://pith.science/pith/EKHB24BRLHG4CC5QH2Q4RWFV75.json","view_paper":"https://pith.science/paper/EKHB24BR","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2505.23683&json=true","fetch_graph":"https://pith.science/api/pith-number/EKHB24BRLHG4CC5QH2Q4RWFV75/graph.json","fetch_events":"https://pith.science/api/pith-number/EKHB24BRLHG4CC5QH2Q4RWFV75/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/EKHB24BRLHG4CC5QH2Q4RWFV75/action/timestamp_anchor","attest_storage":"https://pith.science/pith/EKHB24BRLHG4CC5QH2Q4RWFV75/action/storage_attestation","attest_author":"https://pith.science/pith/EKHB24BRLHG4CC5QH2Q4RWFV75/action/author_attestation","sign_citation":"https://pith.science/pith/EKHB24BRLHG4CC5QH2Q4RWFV75/action/citation_signature","submit_replication":"https://pith.science/pith/EKHB24BRLHG4CC5QH2Q4RWFV75/action/replication_record"}},"created_at":"2026-07-05T11:12:12.700408+00:00","updated_at":"2026-07-05T11:12:12.700408+00:00"}