{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:UVTIJ6BCB3WPBBV44NHJHA5J2Z","short_pith_number":"pith:UVTIJ6BC","schema_version":"1.0","canonical_sha256":"a56684f8220eecf086bce34e9383a9d6586f8aae18b2231349bd758ac227ac1c","source":{"kind":"arxiv","id":"2502.08991","version":2},"attestation_state":"computed","paper":{"title":"Task Generalization With AutoRegressive Compositional Structure: Can Learning From $D$ Tasks Generalize to $D^{T}$ Tasks?","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["stat.ML"],"primary_cat":"cs.LG","authors_text":"Amirhesam Abedsoltan, Hongzhou Lin, Huaqing Zhang, Jingzhao Zhang, Kaiyue Wen, Mikhail Belkin","submitted_at":"2025-02-13T06:08:01Z","abstract_excerpt":"Large language models (LLMs) exhibit remarkable task generalization, solving tasks they were never explicitly trained on with only a few demonstrations. This raises a fundamental question: When can learning from a small set of tasks generalize to a large task family? In this paper, we investigate task generalization through the lens of autoregressive compositional structure, where each task is a composition of $T$ operations, and each operation is among a finite family of $D$ subtasks. This yields a total class of size $D^T$. We first show that generalization to all $D^T$ tasks is theoreticall"},"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":"2502.08991","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2025-02-13T06:08:01Z","cross_cats_sorted":["stat.ML"],"title_canon_sha256":"1597762ca6b304137279c51a256f2c578d43270367a7f17c2f5c7c18295f3538","abstract_canon_sha256":"ec3ea5b05f4da79019795c4cb781dfe6e7fd16b8dba84f66e6f494ab98e247d1"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:18:17.954778Z","signature_b64":"oveA4k2cDCtlliqyoNBKMJBa+kQOF9tKT8pb+GWeZVrcG3uhE/XuHwy9Z+v4q7fxNgTAkRpxLDsDqfJUtg2RDQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"a56684f8220eecf086bce34e9383a9d6586f8aae18b2231349bd758ac227ac1c","last_reissued_at":"2026-07-05T11:18:17.954278Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:18:17.954278Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Task Generalization With AutoRegressive Compositional Structure: Can Learning From $D$ Tasks Generalize to $D^{T}$ Tasks?","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["stat.ML"],"primary_cat":"cs.LG","authors_text":"Amirhesam Abedsoltan, Hongzhou Lin, Huaqing Zhang, Jingzhao Zhang, Kaiyue Wen, Mikhail Belkin","submitted_at":"2025-02-13T06:08:01Z","abstract_excerpt":"Large language models (LLMs) exhibit remarkable task generalization, solving tasks they were never explicitly trained on with only a few demonstrations. This raises a fundamental question: When can learning from a small set of tasks generalize to a large task family? In this paper, we investigate task generalization through the lens of autoregressive compositional structure, where each task is a composition of $T$ operations, and each operation is among a finite family of $D$ subtasks. This yields a total class of size $D^T$. We first show that generalization to all $D^T$ tasks is theoreticall"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2502.08991","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/2502.08991/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":"2502.08991","created_at":"2026-07-05T11:18:17.954356+00:00"},{"alias_kind":"arxiv_version","alias_value":"2502.08991v2","created_at":"2026-07-05T11:18:17.954356+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2502.08991","created_at":"2026-07-05T11:18:17.954356+00:00"},{"alias_kind":"pith_short_12","alias_value":"UVTIJ6BCB3WP","created_at":"2026-07-05T11:18:17.954356+00:00"},{"alias_kind":"pith_short_16","alias_value":"UVTIJ6BCB3WPBBV4","created_at":"2026-07-05T11:18:17.954356+00:00"},{"alias_kind":"pith_short_8","alias_value":"UVTIJ6BC","created_at":"2026-07-05T11:18:17.954356+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":2,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.18089","citing_title":"From Reasoning Traces to Reusable Modules: Understanding Compositional Generalization in Language Model Reasoning","ref_index":145,"is_internal_anchor":false},{"citing_arxiv_id":"2604.15306","citing_title":"Generalization in LLM Problem Solving: The Case of the Shortest Path","ref_index":1,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/UVTIJ6BCB3WPBBV44NHJHA5J2Z","json":"https://pith.science/pith/UVTIJ6BCB3WPBBV44NHJHA5J2Z.json","graph_json":"https://pith.science/api/pith-number/UVTIJ6BCB3WPBBV44NHJHA5J2Z/graph.json","events_json":"https://pith.science/api/pith-number/UVTIJ6BCB3WPBBV44NHJHA5J2Z/events.json","paper":"https://pith.science/paper/UVTIJ6BC"},"agent_actions":{"view_html":"https://pith.science/pith/UVTIJ6BCB3WPBBV44NHJHA5J2Z","download_json":"https://pith.science/pith/UVTIJ6BCB3WPBBV44NHJHA5J2Z.json","view_paper":"https://pith.science/paper/UVTIJ6BC","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2502.08991&json=true","fetch_graph":"https://pith.science/api/pith-number/UVTIJ6BCB3WPBBV44NHJHA5J2Z/graph.json","fetch_events":"https://pith.science/api/pith-number/UVTIJ6BCB3WPBBV44NHJHA5J2Z/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/UVTIJ6BCB3WPBBV44NHJHA5J2Z/action/timestamp_anchor","attest_storage":"https://pith.science/pith/UVTIJ6BCB3WPBBV44NHJHA5J2Z/action/storage_attestation","attest_author":"https://pith.science/pith/UVTIJ6BCB3WPBBV44NHJHA5J2Z/action/author_attestation","sign_citation":"https://pith.science/pith/UVTIJ6BCB3WPBBV44NHJHA5J2Z/action/citation_signature","submit_replication":"https://pith.science/pith/UVTIJ6BCB3WPBBV44NHJHA5J2Z/action/replication_record"}},"created_at":"2026-07-05T11:18:17.954356+00:00","updated_at":"2026-07-05T11:18:17.954356+00:00"}