{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2021:AZS6CKZHL3QGAJNA6PLQV2WPRJ","short_pith_number":"pith:AZS6CKZH","schema_version":"1.0","canonical_sha256":"0665e12b275ee06025a0f3d70aeacf8a401527175e044ae49f3b2c042a66b54b","source":{"kind":"arxiv","id":"2111.04850","version":3},"attestation_state":"computed","paper":{"title":"Dueling RL: Reinforcement Learning with Trajectory Preferences","license":"http://creativecommons.org/publicdomain/zero/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Aadirupa Saha, Aldo Pacchiano, Jonathan Lee","submitted_at":"2021-11-08T22:17:36Z","abstract_excerpt":"We consider the problem of preference based reinforcement learning (PbRL), where, unlike traditional reinforcement learning, an agent receives feedback only in terms of a 1 bit (0/1) preference over a trajectory pair instead of absolute rewards for them. The success of the traditional RL framework crucially relies on the underlying agent-reward model, which, however, depends on how accurately a system designer can express an appropriate reward function and often a non-trivial task. The main novelty of our framework is the ability to learn from preference-based trajectory feedback that eliminat"},"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":"2111.04850","kind":"arxiv","version":3},"metadata":{"license":"http://creativecommons.org/publicdomain/zero/1.0/","primary_cat":"cs.LG","submitted_at":"2021-11-08T22:17:36Z","cross_cats_sorted":[],"title_canon_sha256":"dbe419cdf3a0f6e65760e15abf3b62a5eae7be276ee0e4da5f48574b45e67282","abstract_canon_sha256":"e60ebd6cb9ef0a4abcce9ef23e96b50030b555c907b2ffdbe794a19673e3c708"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T05:39:00.550796Z","signature_b64":"/RuVRTTF6sGFVuNjqz3u8TxhQI6GRXCvn5lorNd8ylz4MP810A0tlHS4ETW24K2N5Wgo/sz3muAx4O8crsioDQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"0665e12b275ee06025a0f3d70aeacf8a401527175e044ae49f3b2c042a66b54b","last_reissued_at":"2026-07-05T05:39:00.550402Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T05:39:00.550402Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Dueling RL: Reinforcement Learning with Trajectory Preferences","license":"http://creativecommons.org/publicdomain/zero/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Aadirupa Saha, Aldo Pacchiano, Jonathan Lee","submitted_at":"2021-11-08T22:17:36Z","abstract_excerpt":"We consider the problem of preference based reinforcement learning (PbRL), where, unlike traditional reinforcement learning, an agent receives feedback only in terms of a 1 bit (0/1) preference over a trajectory pair instead of absolute rewards for them. The success of the traditional RL framework crucially relies on the underlying agent-reward model, which, however, depends on how accurately a system designer can express an appropriate reward function and often a non-trivial task. The main novelty of our framework is the ability to learn from preference-based trajectory feedback that eliminat"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2111.04850","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/2111.04850/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":"2111.04850","created_at":"2026-07-05T05:39:00.550461+00:00"},{"alias_kind":"arxiv_version","alias_value":"2111.04850v3","created_at":"2026-07-05T05:39:00.550461+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2111.04850","created_at":"2026-07-05T05:39:00.550461+00:00"},{"alias_kind":"pith_short_12","alias_value":"AZS6CKZHL3QG","created_at":"2026-07-05T05:39:00.550461+00:00"},{"alias_kind":"pith_short_16","alias_value":"AZS6CKZHL3QGAJNA","created_at":"2026-07-05T05:39:00.550461+00:00"},{"alias_kind":"pith_short_8","alias_value":"AZS6CKZH","created_at":"2026-07-05T05:39:00.550461+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":6,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.27082","citing_title":"Finding Stationary Points by Comparisons","ref_index":41,"is_internal_anchor":false},{"citing_arxiv_id":"2607.01925","citing_title":"SPLC: Social Preference Learning for Crowd Robot Navigation","ref_index":31,"is_internal_anchor":false},{"citing_arxiv_id":"2606.06053","citing_title":"Online KL-Regularized Reinforcement Learning with Function Approximation under Misspecification","ref_index":11,"is_internal_anchor":false},{"citing_arxiv_id":"2606.30271","citing_title":"Preference-Based Reward Learning under Partial Observability with Inexact Dynamics","ref_index":14,"is_internal_anchor":false},{"citing_arxiv_id":"2510.08539","citing_title":"On the optimization dynamics of RLVR: Gradient gap and step size thresholds","ref_index":15,"is_internal_anchor":false},{"citing_arxiv_id":"2604.02349","citing_title":"OPRIDE: Offline Preference-based Reinforcement Learning via In-Dataset Exploration","ref_index":36,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/AZS6CKZHL3QGAJNA6PLQV2WPRJ","json":"https://pith.science/pith/AZS6CKZHL3QGAJNA6PLQV2WPRJ.json","graph_json":"https://pith.science/api/pith-number/AZS6CKZHL3QGAJNA6PLQV2WPRJ/graph.json","events_json":"https://pith.science/api/pith-number/AZS6CKZHL3QGAJNA6PLQV2WPRJ/events.json","paper":"https://pith.science/paper/AZS6CKZH"},"agent_actions":{"view_html":"https://pith.science/pith/AZS6CKZHL3QGAJNA6PLQV2WPRJ","download_json":"https://pith.science/pith/AZS6CKZHL3QGAJNA6PLQV2WPRJ.json","view_paper":"https://pith.science/paper/AZS6CKZH","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2111.04850&json=true","fetch_graph":"https://pith.science/api/pith-number/AZS6CKZHL3QGAJNA6PLQV2WPRJ/graph.json","fetch_events":"https://pith.science/api/pith-number/AZS6CKZHL3QGAJNA6PLQV2WPRJ/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/AZS6CKZHL3QGAJNA6PLQV2WPRJ/action/timestamp_anchor","attest_storage":"https://pith.science/pith/AZS6CKZHL3QGAJNA6PLQV2WPRJ/action/storage_attestation","attest_author":"https://pith.science/pith/AZS6CKZHL3QGAJNA6PLQV2WPRJ/action/author_attestation","sign_citation":"https://pith.science/pith/AZS6CKZHL3QGAJNA6PLQV2WPRJ/action/citation_signature","submit_replication":"https://pith.science/pith/AZS6CKZHL3QGAJNA6PLQV2WPRJ/action/replication_record"}},"created_at":"2026-07-05T05:39:00.550461+00:00","updated_at":"2026-07-05T05:39:00.550461+00:00"}