{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:SVXXI4RIQ7KUZA5HU44OOASDE4","short_pith_number":"pith:SVXXI4RI","schema_version":"1.0","canonical_sha256":"956f74722887d54c83a7a738e70243272d70121fb2f6fe642bdba285a1bb4db5","source":{"kind":"arxiv","id":"2506.15446","version":1},"attestation_state":"computed","paper":{"title":"Zero-Shot Reinforcement Learning Under Partial Observability","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Jonathan M. Cullen, Scott Jeen, Tom Bewley","submitted_at":"2025-06-18T13:18:36Z","abstract_excerpt":"Recent work has shown that, under certain assumptions, zero-shot reinforcement learning (RL) methods can generalise to any unseen task in an environment after reward-free pre-training. Access to Markov states is one such assumption, yet, in many real-world applications, the Markov state is only partially observable. Here, we explore how the performance of standard zero-shot RL methods degrades when subjected to partially observability, and show that, as in single-task RL, memory-based architectures are an effective remedy. We evaluate our memory-based zero-shot RL methods in domains where the "},"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":"2506.15446","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2025-06-18T13:18:36Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"1e790a721b3806224733327cabed0d31f602746fecdf50ec32709768cdd242dd","abstract_canon_sha256":"b1e2b9a28aee2532bbb2140c7f884cc4e4a1f862d25a9ac14d4bbe079d8ba0fe"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:23:43.678468Z","signature_b64":"yNcjCP3JV8BCkD+esox9Fx82fJ6aNJeTDT97aPWNYFNmdBV3jOP7KuGM9Fdm7eN+Ov2aUgCAmi0VAnqK6gHdDA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"956f74722887d54c83a7a738e70243272d70121fb2f6fe642bdba285a1bb4db5","last_reissued_at":"2026-07-05T11:23:43.677989Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:23:43.677989Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Zero-Shot Reinforcement Learning Under Partial Observability","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Jonathan M. Cullen, Scott Jeen, Tom Bewley","submitted_at":"2025-06-18T13:18:36Z","abstract_excerpt":"Recent work has shown that, under certain assumptions, zero-shot reinforcement learning (RL) methods can generalise to any unseen task in an environment after reward-free pre-training. Access to Markov states is one such assumption, yet, in many real-world applications, the Markov state is only partially observable. Here, we explore how the performance of standard zero-shot RL methods degrades when subjected to partially observability, and show that, as in single-task RL, memory-based architectures are an effective remedy. We evaluate our memory-based zero-shot RL methods in domains where the "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2506.15446","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/2506.15446/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":"2506.15446","created_at":"2026-07-05T11:23:43.678045+00:00"},{"alias_kind":"arxiv_version","alias_value":"2506.15446v1","created_at":"2026-07-05T11:23:43.678045+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2506.15446","created_at":"2026-07-05T11:23:43.678045+00:00"},{"alias_kind":"pith_short_12","alias_value":"SVXXI4RIQ7KU","created_at":"2026-07-05T11:23:43.678045+00:00"},{"alias_kind":"pith_short_16","alias_value":"SVXXI4RIQ7KUZA5H","created_at":"2026-07-05T11:23:43.678045+00:00"},{"alias_kind":"pith_short_8","alias_value":"SVXXI4RI","created_at":"2026-07-05T11:23:43.678045+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2605.17017","citing_title":"When Dynamics Shift, Robust Task Inference Wins: Offline Imitation Learning with Behavior Foundation Models Revisited","ref_index":30,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/SVXXI4RIQ7KUZA5HU44OOASDE4","json":"https://pith.science/pith/SVXXI4RIQ7KUZA5HU44OOASDE4.json","graph_json":"https://pith.science/api/pith-number/SVXXI4RIQ7KUZA5HU44OOASDE4/graph.json","events_json":"https://pith.science/api/pith-number/SVXXI4RIQ7KUZA5HU44OOASDE4/events.json","paper":"https://pith.science/paper/SVXXI4RI"},"agent_actions":{"view_html":"https://pith.science/pith/SVXXI4RIQ7KUZA5HU44OOASDE4","download_json":"https://pith.science/pith/SVXXI4RIQ7KUZA5HU44OOASDE4.json","view_paper":"https://pith.science/paper/SVXXI4RI","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2506.15446&json=true","fetch_graph":"https://pith.science/api/pith-number/SVXXI4RIQ7KUZA5HU44OOASDE4/graph.json","fetch_events":"https://pith.science/api/pith-number/SVXXI4RIQ7KUZA5HU44OOASDE4/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/SVXXI4RIQ7KUZA5HU44OOASDE4/action/timestamp_anchor","attest_storage":"https://pith.science/pith/SVXXI4RIQ7KUZA5HU44OOASDE4/action/storage_attestation","attest_author":"https://pith.science/pith/SVXXI4RIQ7KUZA5HU44OOASDE4/action/author_attestation","sign_citation":"https://pith.science/pith/SVXXI4RIQ7KUZA5HU44OOASDE4/action/citation_signature","submit_replication":"https://pith.science/pith/SVXXI4RIQ7KUZA5HU44OOASDE4/action/replication_record"}},"created_at":"2026-07-05T11:23:43.678045+00:00","updated_at":"2026-07-05T11:23:43.678045+00:00"}