{"paper":{"title":"SOPE: Stabilizing Off-Policy Evaluation for Online RL with Prior Data","license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","headline":"SOPE uses an actor-aligned OPE signal on held-out data to automatically stop offline phases in online RL with prior data.","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Alessandro Sestini, Andrew D. Bagdanov, Carlo Romeo, Girolamo Macaluso","submitted_at":"2026-05-07T08:32:09Z","abstract_excerpt":"Incorporating prior data into online reinforcement learning accelerates training but typically forces a difficult trade-off between high computational costs and long, multi-stage training pipelines. While fixed-length stabilization phases are significantly more computationally efficient than static update schedules, they require task-dependent manual tuning, risking either the waste of prior knowledge or severe overfitting. To address this, we propose SOPE, an algorithm that uses an actor-aligned Off-Policy Policy Evaluation (OPE) signal as an automated early-stopping mechanism to dynamically "},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"Evaluated on 25 continuous control tasks from the Minari benchmark suite, SOPE improves baseline performance by up to 45.6% while reducing the required TFLOPs by up to 22x, thus balancing the tradeoff between sample and computational efficiency.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That the actor-aligned OPE signal evaluated on a held-out validation split under the current policy's action distribution accurately detects the saturation point of out-of-distribution benefits without either stopping too early (wasting prior knowledge) or too late (causing overfitting).","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"SOPE uses an actor-aligned OPE signal on a held-out validation split to dynamically stop offline stabilization phases in online RL, improving performance up to 45.6% and cutting TFLOPs up to 22x on 25 Minari tasks.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"SOPE uses an actor-aligned OPE signal on held-out data to automatically stop offline phases in online RL with prior data.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"9d9f3a58267f272a031f4a8ec3df5db7e734dda18b4f217eba9b3e2c990741e3"},"source":{"id":"2605.05863","kind":"arxiv","version":2},"verdict":{"id":"cbbfe96c-6794-43dd-8ec4-ae751b735dc1","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-08T14:50:37.993640Z","strongest_claim":"Evaluated on 25 continuous control tasks from the Minari benchmark suite, SOPE improves baseline performance by up to 45.6% while reducing the required TFLOPs by up to 22x, thus balancing the tradeoff between sample and computational efficiency.","one_line_summary":"SOPE uses an actor-aligned OPE signal on a held-out validation split to dynamically stop offline stabilization phases in online RL, improving performance up to 45.6% and cutting TFLOPs up to 22x on 25 Minari tasks.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That the actor-aligned OPE signal evaluated on a held-out validation split under the current policy's action distribution accurately detects the saturation point of out-of-distribution benefits without either stopping too early (wasting prior knowledge) or too late (causing overfitting).","pith_extraction_headline":"SOPE uses an actor-aligned OPE signal on held-out data to automatically stop offline phases in online RL with prior data."},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2605.05863/integrity.json","findings":[],"available":true,"detectors_run":[{"name":"claim_evidence","ran_at":"2026-05-20T13:42:04.568842Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"ai_meta_artifact","ran_at":"2026-05-20T08:41:37.757500Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"doi_title_agreement","ran_at":"2026-05-19T19:31:19.547150Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"doi_compliance","ran_at":"2026-05-19T13:10:19.048833Z","status":"completed","version":"1.0.0","findings_count":0}],"snapshot_sha256":"b18319b5272c7d5c096da65596b096e7694a12c061693846f15a8f159d8db496"},"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"}