{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:37I6P3OYFCFSMKVYT4QEWIHKXJ","short_pith_number":"pith:37I6P3OY","schema_version":"1.0","canonical_sha256":"dfd1e7edd8288b262ab89f204b20eaba60b4be89fbb853c4414a201d6c3e11eb","source":{"kind":"arxiv","id":"2511.17852","version":2},"attestation_state":"computed","paper":{"title":"Transformers with RL or SFT Provably Learn Sparse Boolean Functions, But Differently","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["stat.ML"],"primary_cat":"cs.LG","authors_text":"Bochen Lyu, Xiaohao Cai, Yiyang Jia, Zhanxing Zhu","submitted_at":"2025-11-22T00:38:43Z","abstract_excerpt":"Transformers can acquire Chain-of-Thought (CoT) capabilities to solve complex reasoning tasks through fine-tuning. Reinforcement learning (RL) and supervised fine-tuning (SFT) are two primary approaches to this end. In this work, we specifically examine RL with process rewards and SFT for learning $k$-sparse Boolean functions with a one-layer transformer through intermediate reasoning steps akin to CoT. In particular, we consider $k$-sparse Boolean functions that can be recursively decomposed into fixed 2-sparse Boolean functions. We first analyze the learning dynamics of RL fine-tuning with p"},"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":"2511.17852","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2025-11-22T00:38:43Z","cross_cats_sorted":["stat.ML"],"title_canon_sha256":"2be2e0a131213221a17634eb3b2e2d755bd5ce2630eeaf003946191e42fbacad","abstract_canon_sha256":"83571253e79a41fa8046eca2c4a7fd5b7c053d7d0db4c60500a9bbbe54a90c74"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-27T01:04:52.702947Z","signature_b64":"5kq7+tSpDNVMAzjMTHzxF0wd7TE3k2XFbtUEuAiTokFup1ZI8f/pKiN+zPdO/dD+rOODXw21sdmjOY7dZPUeAg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"dfd1e7edd8288b262ab89f204b20eaba60b4be89fbb853c4414a201d6c3e11eb","last_reissued_at":"2026-05-27T01:04:52.701750Z","signature_status":"signed_v1","first_computed_at":"2026-05-27T01:04:52.701750Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Transformers with RL or SFT Provably Learn Sparse Boolean Functions, But Differently","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["stat.ML"],"primary_cat":"cs.LG","authors_text":"Bochen Lyu, Xiaohao Cai, Yiyang Jia, Zhanxing Zhu","submitted_at":"2025-11-22T00:38:43Z","abstract_excerpt":"Transformers can acquire Chain-of-Thought (CoT) capabilities to solve complex reasoning tasks through fine-tuning. Reinforcement learning (RL) and supervised fine-tuning (SFT) are two primary approaches to this end. In this work, we specifically examine RL with process rewards and SFT for learning $k$-sparse Boolean functions with a one-layer transformer through intermediate reasoning steps akin to CoT. In particular, we consider $k$-sparse Boolean functions that can be recursively decomposed into fixed 2-sparse Boolean functions. We first analyze the learning dynamics of RL fine-tuning with p"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2511.17852","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/2511.17852/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":"2511.17852","created_at":"2026-05-27T01:04:52.701911+00:00"},{"alias_kind":"arxiv_version","alias_value":"2511.17852v2","created_at":"2026-05-27T01:04:52.701911+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2511.17852","created_at":"2026-05-27T01:04:52.701911+00:00"},{"alias_kind":"pith_short_12","alias_value":"37I6P3OYFCFS","created_at":"2026-05-27T01:04:52.701911+00:00"},{"alias_kind":"pith_short_16","alias_value":"37I6P3OYFCFSMKVY","created_at":"2026-05-27T01:04:52.701911+00:00"},{"alias_kind":"pith_short_8","alias_value":"37I6P3OY","created_at":"2026-05-27T01:04:52.701911+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":0,"internal_anchor_count":0,"sample":[]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/37I6P3OYFCFSMKVYT4QEWIHKXJ","json":"https://pith.science/pith/37I6P3OYFCFSMKVYT4QEWIHKXJ.json","graph_json":"https://pith.science/api/pith-number/37I6P3OYFCFSMKVYT4QEWIHKXJ/graph.json","events_json":"https://pith.science/api/pith-number/37I6P3OYFCFSMKVYT4QEWIHKXJ/events.json","paper":"https://pith.science/paper/37I6P3OY"},"agent_actions":{"view_html":"https://pith.science/pith/37I6P3OYFCFSMKVYT4QEWIHKXJ","download_json":"https://pith.science/pith/37I6P3OYFCFSMKVYT4QEWIHKXJ.json","view_paper":"https://pith.science/paper/37I6P3OY","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2511.17852&json=true","fetch_graph":"https://pith.science/api/pith-number/37I6P3OYFCFSMKVYT4QEWIHKXJ/graph.json","fetch_events":"https://pith.science/api/pith-number/37I6P3OYFCFSMKVYT4QEWIHKXJ/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/37I6P3OYFCFSMKVYT4QEWIHKXJ/action/timestamp_anchor","attest_storage":"https://pith.science/pith/37I6P3OYFCFSMKVYT4QEWIHKXJ/action/storage_attestation","attest_author":"https://pith.science/pith/37I6P3OYFCFSMKVYT4QEWIHKXJ/action/author_attestation","sign_citation":"https://pith.science/pith/37I6P3OYFCFSMKVYT4QEWIHKXJ/action/citation_signature","submit_replication":"https://pith.science/pith/37I6P3OYFCFSMKVYT4QEWIHKXJ/action/replication_record"}},"created_at":"2026-05-27T01:04:52.701911+00:00","updated_at":"2026-05-27T01:04:52.701911+00:00"}