{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:6DVK2FC5OGAR44MRQPHYUJLEHS","short_pith_number":"pith:6DVK2FC5","schema_version":"1.0","canonical_sha256":"f0eaad145d71811e719183cf8a25643caf8aeeeff2d2e1757e508e941e32305f","source":{"kind":"arxiv","id":"2509.22047","version":3},"attestation_state":"computed","paper":{"title":"MO-GRPO: Mitigating Reward Hacking of Group Relative Policy Optimization on Multi-Objective Problems","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Eiji Uchibe, Mitsuki Sakamoto, Ryota Mitsuhashi, Tetsuro Morimura, Yuki Ichihara, Yuu Jinnai","submitted_at":"2025-09-26T08:32:22Z","abstract_excerpt":"Group Relative Policy Optimization (GRPO) has been shown to be an effective algorithm when an accurate reward model is available. However, such a highly reliable reward model is not available in many real-world tasks. In this paper, we particularly focus on multi-objective settings, in which we identify that GRPO is vulnerable to reward hacking, optimizing only one of the objectives at the cost of the others. To address this issue, we propose MO-GRPO, an extension of GRPO with a simple normalization method to reweight the reward functions automatically according to the variances of their value"},"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":"2509.22047","kind":"arxiv","version":3},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2025-09-26T08:32:22Z","cross_cats_sorted":[],"title_canon_sha256":"304fb937b48591c81854ed2f9fb27f8272b3f074cf673301ca9101ab6443c316","abstract_canon_sha256":"3b74c7364e472fc32b0a569218e22ab7952e1c00617d38fd4f1e2713b5a448ba"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-24T00:23:04.867235Z","signature_b64":"vshfnu2e2Ng35Ct1SAtF4UNdq7XVeelm6xEnnh7+EtSAnoX+/rVBEPsrg07YEXl8Enh5So2MEpAR8ugwSUFVCQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"f0eaad145d71811e719183cf8a25643caf8aeeeff2d2e1757e508e941e32305f","last_reissued_at":"2026-07-24T00:23:04.866259Z","signature_status":"signed_v1","first_computed_at":"2026-07-24T00:23:04.866259Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"MO-GRPO: Mitigating Reward Hacking of Group Relative Policy Optimization on Multi-Objective Problems","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Eiji Uchibe, Mitsuki Sakamoto, Ryota Mitsuhashi, Tetsuro Morimura, Yuki Ichihara, Yuu Jinnai","submitted_at":"2025-09-26T08:32:22Z","abstract_excerpt":"Group Relative Policy Optimization (GRPO) has been shown to be an effective algorithm when an accurate reward model is available. However, such a highly reliable reward model is not available in many real-world tasks. In this paper, we particularly focus on multi-objective settings, in which we identify that GRPO is vulnerable to reward hacking, optimizing only one of the objectives at the cost of the others. To address this issue, we propose MO-GRPO, an extension of GRPO with a simple normalization method to reweight the reward functions automatically according to the variances of their value"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2509.22047","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/2509.22047/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":"2509.22047","created_at":"2026-07-24T00:23:04.866734+00:00"},{"alias_kind":"arxiv_version","alias_value":"2509.22047v3","created_at":"2026-07-24T00:23:04.866734+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2509.22047","created_at":"2026-07-24T00:23:04.866734+00:00"},{"alias_kind":"pith_short_12","alias_value":"6DVK2FC5OGAR","created_at":"2026-07-24T00:23:04.866734+00:00"},{"alias_kind":"pith_short_16","alias_value":"6DVK2FC5OGAR44MR","created_at":"2026-07-24T00:23:04.866734+00:00"},{"alias_kind":"pith_short_8","alias_value":"6DVK2FC5","created_at":"2026-07-24T00:23:04.866734+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":3,"internal_anchor_count":3,"sample":[{"citing_arxiv_id":"2605.24005","citing_title":"LC-ERD: Mining Latent Logic for Self-Evolving Reasoning via Consistency-Regulated Reward Decomposition","ref_index":13,"is_internal_anchor":true},{"citing_arxiv_id":"2606.21413","citing_title":"CAT-Translate: Building Compact Open-Source Models for Japanese-English Translation","ref_index":20,"is_internal_anchor":true},{"citing_arxiv_id":"2605.07276","citing_title":"Signal Reshaping for GRPO in Weak-Feedback Agentic Code Repair","ref_index":10,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/6DVK2FC5OGAR44MRQPHYUJLEHS","json":"https://pith.science/pith/6DVK2FC5OGAR44MRQPHYUJLEHS.json","graph_json":"https://pith.science/api/pith-number/6DVK2FC5OGAR44MRQPHYUJLEHS/graph.json","events_json":"https://pith.science/api/pith-number/6DVK2FC5OGAR44MRQPHYUJLEHS/events.json","paper":"https://pith.science/paper/6DVK2FC5"},"agent_actions":{"view_html":"https://pith.science/pith/6DVK2FC5OGAR44MRQPHYUJLEHS","download_json":"https://pith.science/pith/6DVK2FC5OGAR44MRQPHYUJLEHS.json","view_paper":"https://pith.science/paper/6DVK2FC5","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2509.22047&json=true","fetch_graph":"https://pith.science/api/pith-number/6DVK2FC5OGAR44MRQPHYUJLEHS/graph.json","fetch_events":"https://pith.science/api/pith-number/6DVK2FC5OGAR44MRQPHYUJLEHS/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/6DVK2FC5OGAR44MRQPHYUJLEHS/action/timestamp_anchor","attest_storage":"https://pith.science/pith/6DVK2FC5OGAR44MRQPHYUJLEHS/action/storage_attestation","attest_author":"https://pith.science/pith/6DVK2FC5OGAR44MRQPHYUJLEHS/action/author_attestation","sign_citation":"https://pith.science/pith/6DVK2FC5OGAR44MRQPHYUJLEHS/action/citation_signature","submit_replication":"https://pith.science/pith/6DVK2FC5OGAR44MRQPHYUJLEHS/action/replication_record"}},"created_at":"2026-07-24T00:23:04.866734+00:00","updated_at":"2026-07-24T00:23:04.866734+00:00"}