{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:574LWMSMGCNB5ZVRVGRPQ65D25","short_pith_number":"pith:574LWMSM","schema_version":"1.0","canonical_sha256":"eff8bb324c309a1ee6b1a9a2f87ba3d7662f6f91a89dd256ec1694b80a53c4e4","source":{"kind":"arxiv","id":"2605.24749","version":1},"attestation_state":"computed","paper":{"title":"How Neural Reward Models Learn Features for Policy Optimization: A Single-Index Analysis","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"stat.ML","authors_text":"Akifumi Wachi, Kohei Miyaguchi, Rei Higuchi, Ryotaro Kawata, Shokichi Takakura, Taiji Suzuki","submitted_at":"2026-05-23T22:00:38Z","abstract_excerpt":"Reward modeling is not only a prediction problem: in KL-regularized policy optimization, the learned reward is exponentiated to define the deployed policy, so downstream value depends on errors in reward-tilted regions. We study this feedback in a Gaussian single-index model with $r^*(x) = \\sigma^*(\\langle \\theta^*, x\\rangle)$ and $x \\sim N(0, I_d)$. We analyze a two-stage neural reward model that first learns the hidden direction $\\theta^*$ from reward-weighted samples and then fits the readout layer by weighted ridge regression. Exponential reward weighting changes the Hermite signal availab"},"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":"2605.24749","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"stat.ML","submitted_at":"2026-05-23T22:00:38Z","cross_cats_sorted":["cs.LG"],"title_canon_sha256":"b663e3c6535e6b8f5e2f8b667cca5fdf070daefd5a8c0a0ba0fb65e1593c9d1a","abstract_canon_sha256":"e3a1e8110a18ab3ba1ef4c03967d65cae0732e8299d33dda91259a683fabc5a8"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-26T01:03:56.318506Z","signature_b64":"feNsUihYSNdCqf4BH7S4ZdzGK+ggzLjzkkcLoRWXRR5lrx/+/IJOEWWxeyr+ca8EPzVxpPDtKtGUzTPByrcRAw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"eff8bb324c309a1ee6b1a9a2f87ba3d7662f6f91a89dd256ec1694b80a53c4e4","last_reissued_at":"2026-05-26T01:03:56.317717Z","signature_status":"signed_v1","first_computed_at":"2026-05-26T01:03:56.317717Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"How Neural Reward Models Learn Features for Policy Optimization: A Single-Index Analysis","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"stat.ML","authors_text":"Akifumi Wachi, Kohei Miyaguchi, Rei Higuchi, Ryotaro Kawata, Shokichi Takakura, Taiji Suzuki","submitted_at":"2026-05-23T22:00:38Z","abstract_excerpt":"Reward modeling is not only a prediction problem: in KL-regularized policy optimization, the learned reward is exponentiated to define the deployed policy, so downstream value depends on errors in reward-tilted regions. We study this feedback in a Gaussian single-index model with $r^*(x) = \\sigma^*(\\langle \\theta^*, x\\rangle)$ and $x \\sim N(0, I_d)$. We analyze a two-stage neural reward model that first learns the hidden direction $\\theta^*$ from reward-weighted samples and then fits the readout layer by weighted ridge regression. Exponential reward weighting changes the Hermite signal availab"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2605.24749","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/2605.24749/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":"2605.24749","created_at":"2026-05-26T01:03:56.317819+00:00"},{"alias_kind":"arxiv_version","alias_value":"2605.24749v1","created_at":"2026-05-26T01:03:56.317819+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.24749","created_at":"2026-05-26T01:03:56.317819+00:00"},{"alias_kind":"pith_short_12","alias_value":"574LWMSMGCNB","created_at":"2026-05-26T01:03:56.317819+00:00"},{"alias_kind":"pith_short_16","alias_value":"574LWMSMGCNB5ZVR","created_at":"2026-05-26T01:03:56.317819+00:00"},{"alias_kind":"pith_short_8","alias_value":"574LWMSM","created_at":"2026-05-26T01:03:56.317819+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/574LWMSMGCNB5ZVRVGRPQ65D25","json":"https://pith.science/pith/574LWMSMGCNB5ZVRVGRPQ65D25.json","graph_json":"https://pith.science/api/pith-number/574LWMSMGCNB5ZVRVGRPQ65D25/graph.json","events_json":"https://pith.science/api/pith-number/574LWMSMGCNB5ZVRVGRPQ65D25/events.json","paper":"https://pith.science/paper/574LWMSM"},"agent_actions":{"view_html":"https://pith.science/pith/574LWMSMGCNB5ZVRVGRPQ65D25","download_json":"https://pith.science/pith/574LWMSMGCNB5ZVRVGRPQ65D25.json","view_paper":"https://pith.science/paper/574LWMSM","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2605.24749&json=true","fetch_graph":"https://pith.science/api/pith-number/574LWMSMGCNB5ZVRVGRPQ65D25/graph.json","fetch_events":"https://pith.science/api/pith-number/574LWMSMGCNB5ZVRVGRPQ65D25/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/574LWMSMGCNB5ZVRVGRPQ65D25/action/timestamp_anchor","attest_storage":"https://pith.science/pith/574LWMSMGCNB5ZVRVGRPQ65D25/action/storage_attestation","attest_author":"https://pith.science/pith/574LWMSMGCNB5ZVRVGRPQ65D25/action/author_attestation","sign_citation":"https://pith.science/pith/574LWMSMGCNB5ZVRVGRPQ65D25/action/citation_signature","submit_replication":"https://pith.science/pith/574LWMSMGCNB5ZVRVGRPQ65D25/action/replication_record"}},"created_at":"2026-05-26T01:03:56.317819+00:00","updated_at":"2026-05-26T01:03:56.317819+00:00"}