{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:VKWBDRDN5V43BYDFJ7IVWOU4R3","short_pith_number":"pith:VKWBDRDN","schema_version":"1.0","canonical_sha256":"aaac11c46ded79b0e0654fd15b3a9c8ec977213dd016df9d9bf489cd7f472a91","source":{"kind":"arxiv","id":"2504.05812","version":3},"attestation_state":"computed","paper":{"title":"Right Question is Already Half the Answer: Fully Unsupervised LLM Reasoning Incentivization","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Changqing Zhang, Haitao Wu, Peilin Zhao, Qingyang Zhang, Yatao Bian","submitted_at":"2025-04-08T08:48:51Z","abstract_excerpt":"Existing methods to enhance the reasoning capability of large language models predominantly rely on supervised fine-tuning (SFT) followed by reinforcement learning (RL) on reasoning-specific data. These approaches critically depend on external supervisions--such as labeled reasoning traces, verified golden answers, or pre-trained reward models. In this work, we propose Entropy Minimized Policy Optimization (\\ours), which makes an early attempt at fully unsupervised LLM reasoning incentivization. By continuously minimizing the predictive entropy of LLMs on unlabeled questions in a latent semant"},"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":"2504.05812","kind":"arxiv","version":3},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2025-04-08T08:48:51Z","cross_cats_sorted":[],"title_canon_sha256":"bd57732c67e4d439643093817238492cf7391781e935ec675a9a893a235f0b02","abstract_canon_sha256":"967f54c79c1b628e649fd1eefbe579464bda967aae2e80236b6416fd01b57444"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:04:48.741539Z","signature_b64":"Qy4hP2aN8DO+caCQOmmc1eTm73nZNOZm+tRU9dXyh5ELm1xA36R4p9lLr88IH2oHTjSnrCgLMiVBJeCGESmaAA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"aaac11c46ded79b0e0654fd15b3a9c8ec977213dd016df9d9bf489cd7f472a91","last_reissued_at":"2026-07-05T11:04:48.740791Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:04:48.740791Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Right Question is Already Half the Answer: Fully Unsupervised LLM Reasoning Incentivization","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Changqing Zhang, Haitao Wu, Peilin Zhao, Qingyang Zhang, Yatao Bian","submitted_at":"2025-04-08T08:48:51Z","abstract_excerpt":"Existing methods to enhance the reasoning capability of large language models predominantly rely on supervised fine-tuning (SFT) followed by reinforcement learning (RL) on reasoning-specific data. These approaches critically depend on external supervisions--such as labeled reasoning traces, verified golden answers, or pre-trained reward models. In this work, we propose Entropy Minimized Policy Optimization (\\ours), which makes an early attempt at fully unsupervised LLM reasoning incentivization. By continuously minimizing the predictive entropy of LLMs on unlabeled questions in a latent semant"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2504.05812","kind":"arxiv","version":3},"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/2504.05812/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":"2504.05812","created_at":"2026-07-05T11:04:48.740861+00:00"},{"alias_kind":"arxiv_version","alias_value":"2504.05812v3","created_at":"2026-07-05T11:04:48.740861+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2504.05812","created_at":"2026-07-05T11:04:48.740861+00:00"},{"alias_kind":"pith_short_12","alias_value":"VKWBDRDN5V43","created_at":"2026-07-05T11:04:48.740861+00:00"},{"alias_kind":"pith_short_16","alias_value":"VKWBDRDN5V43BYDF","created_at":"2026-07-05T11:04:48.740861+00:00"},{"alias_kind":"pith_short_8","alias_value":"VKWBDRDN","created_at":"2026-07-05T11:04:48.740861+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":19,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2607.08164","citing_title":"Continual Test-Time Adaptation in Computer Vision: Methods, Benchmarks, and Future Directions","ref_index":33,"is_internal_anchor":true},{"citing_arxiv_id":"2606.08501","citing_title":"Back on Track: Aligning Rewards and States for Reasoning in Diffusion Large Language Models","ref_index":44,"is_internal_anchor":false},{"citing_arxiv_id":"2606.01249","citing_title":"Trust Region On-Policy Distillation","ref_index":67,"is_internal_anchor":false},{"citing_arxiv_id":"2605.25864","citing_title":"When Self-Belief Misleads: Active Label Acquisition for Reinforcement Learning with Verifiable Rewards","ref_index":50,"is_internal_anchor":false},{"citing_arxiv_id":"2606.01075","citing_title":"On the Generalization Gap in Self-Evolving Language Model Reasoning","ref_index":46,"is_internal_anchor":false},{"citing_arxiv_id":"2602.12579","citing_title":"VI-CuRL: Stabilizing Verifier-Independent RL Reasoning via Confidence-Guided Variance Reduction","ref_index":36,"is_internal_anchor":false},{"citing_arxiv_id":"2605.21801","citing_title":"Why Semantic Entropy Fails: Geometry-Aware and Calibrated Uncertainty for Policy Optimization","ref_index":18,"is_internal_anchor":false},{"citing_arxiv_id":"2605.11461","citing_title":"Breaking $\\textit{Winner-Takes-All}$: Cooperative Policy Optimization Improves Diverse LLM Reasoning","ref_index":41,"is_internal_anchor":false},{"citing_arxiv_id":"2605.17187","citing_title":"PluRule: A Benchmark for Moderating Pluralistic Communities on Social Media","ref_index":50,"is_internal_anchor":false},{"citing_arxiv_id":"2505.15134","citing_title":"The Unreasonable Effectiveness of Entropy Minimization in LLM Reasoning","ref_index":99,"is_internal_anchor":false},{"citing_arxiv_id":"2504.20571","citing_title":"Reinforcement Learning for Reasoning in Large Language Models with One Training Example","ref_index":46,"is_internal_anchor":false},{"citing_arxiv_id":"2603.27977","citing_title":"SARL: Label-Free Reinforcement Learning by Rewarding Reasoning Topology","ref_index":30,"is_internal_anchor":false},{"citing_arxiv_id":"2604.03993","citing_title":"Can LLMs Learn to Reason Robustly under Noisy Supervision?","ref_index":34,"is_internal_anchor":false},{"citing_arxiv_id":"2605.11461","citing_title":"Breaking $\\textit{Winner-Takes-All}$: Cooperative Policy Optimization Improves Diverse LLM Reasoning","ref_index":41,"is_internal_anchor":false},{"citing_arxiv_id":"2605.08516","citing_title":"OracleTSC: Oracle-Informed Reward Hurdle and Uncertainty Regularization for Traffic Signal Control","ref_index":22,"is_internal_anchor":false},{"citing_arxiv_id":"2605.06642","citing_title":"StraTA: Incentivizing Agentic Reinforcement Learning with Strategic Trajectory Abstraction","ref_index":45,"is_internal_anchor":false},{"citing_arxiv_id":"2604.11547","citing_title":"Eliciting Medical Reasoning with Knowledge-enhanced Data Synthesis: A Semi-Supervised Reinforcement Learning Approach","ref_index":8,"is_internal_anchor":false},{"citing_arxiv_id":"2604.07864","citing_title":"ZeroCoder: Can LLMs Improve Code Generation Without Ground-Truth Supervision?","ref_index":51,"is_internal_anchor":false},{"citing_arxiv_id":"2604.18493","citing_title":"Too Correct to Learn: Reinforcement Learning on Saturated Reasoning Data","ref_index":16,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/VKWBDRDN5V43BYDFJ7IVWOU4R3","json":"https://pith.science/pith/VKWBDRDN5V43BYDFJ7IVWOU4R3.json","graph_json":"https://pith.science/api/pith-number/VKWBDRDN5V43BYDFJ7IVWOU4R3/graph.json","events_json":"https://pith.science/api/pith-number/VKWBDRDN5V43BYDFJ7IVWOU4R3/events.json","paper":"https://pith.science/paper/VKWBDRDN"},"agent_actions":{"view_html":"https://pith.science/pith/VKWBDRDN5V43BYDFJ7IVWOU4R3","download_json":"https://pith.science/pith/VKWBDRDN5V43BYDFJ7IVWOU4R3.json","view_paper":"https://pith.science/paper/VKWBDRDN","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2504.05812&json=true","fetch_graph":"https://pith.science/api/pith-number/VKWBDRDN5V43BYDFJ7IVWOU4R3/graph.json","fetch_events":"https://pith.science/api/pith-number/VKWBDRDN5V43BYDFJ7IVWOU4R3/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/VKWBDRDN5V43BYDFJ7IVWOU4R3/action/timestamp_anchor","attest_storage":"https://pith.science/pith/VKWBDRDN5V43BYDFJ7IVWOU4R3/action/storage_attestation","attest_author":"https://pith.science/pith/VKWBDRDN5V43BYDFJ7IVWOU4R3/action/author_attestation","sign_citation":"https://pith.science/pith/VKWBDRDN5V43BYDFJ7IVWOU4R3/action/citation_signature","submit_replication":"https://pith.science/pith/VKWBDRDN5V43BYDFJ7IVWOU4R3/action/replication_record"}},"created_at":"2026-07-05T11:04:48.740861+00:00","updated_at":"2026-07-05T11:04:48.740861+00:00"}