{"paper":{"title":"Split the Differences, Pool the Rest: Provably Efficient Multi-Objective Imitation","license":"http://creativecommons.org/licenses/by/4.0/","headline":"MA-BC partitions conflicting expert data and pools the rest to recover Pareto-optimal policies faster than separate learners in multi-objective imitation.","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Claire Vernade, Luca Viano, Volkan Cevher, Ziyad Sheebaelhamd","submitted_at":"2026-05-12T11:49:08Z","abstract_excerpt":"This work investigates multi-objective imitation learning: the problem of recovering policies that lie on the Pareto front given demonstrations from multiple Pareto-optimal experts in a Multi-Objective Markov Decision Process (MOMDP). Standard imitation approaches are ill-equipped for this regime, as naively aggregating conflicting expert trajectories can result in dominated policies. To address this, we introduce Multi-Output Augmented Behavioral Cloning (MA-BC), an algorithm that systematically partitions divergent expert data while pooling state-action pairs where no behavior conflict is ob"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"MA-BC converges to Pareto-optimal policies at a faster statistical rate than any learner that considers each expert dataset independently, and is minimax optimal.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"The provided demonstrations come from Pareto-optimal experts in a MOMDP, and that observable conflicts in state-action pairs can be reliably partitioned without additional structure on the transition dynamics or reward functions.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"MA-BC partitions divergent expert data while pooling non-conflicting pairs in MOMDPs, converging faster to Pareto-optimal policies than independent learners and matching a new minimax lower bound.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"MA-BC partitions conflicting expert data and pools the rest to recover Pareto-optimal policies faster than separate learners in multi-objective imitation.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"aa6fca9d1b225c6a44f2a7ab877d9b7edf23f5e6a39790128c8929ded3399781"},"source":{"id":"2605.12000","kind":"arxiv","version":2},"verdict":{"id":"883292d0-26f6-45c3-ade5-7f2e115f4577","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-13T07:12:17.881078Z","strongest_claim":"MA-BC converges to Pareto-optimal policies at a faster statistical rate than any learner that considers each expert dataset independently, and is minimax optimal.","one_line_summary":"MA-BC partitions divergent expert data while pooling non-conflicting pairs in MOMDPs, converging faster to Pareto-optimal policies than independent learners and matching a new minimax lower bound.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"The provided demonstrations come from Pareto-optimal experts in a MOMDP, and that observable conflicts in state-action pairs can be reliably partitioned without additional structure on the transition dynamics or reward functions.","pith_extraction_headline":"MA-BC partitions conflicting expert data and pools the rest to recover Pareto-optimal policies faster than separate learners in multi-objective imitation."},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2605.12000/integrity.json","findings":[],"available":true,"detectors_run":[{"name":"ai_meta_artifact","ran_at":"2026-05-19T11:34:32.698951Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"doi_title_agreement","ran_at":"2026-05-19T09:01:16.975212Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"doi_compliance","ran_at":"2026-05-19T07:59:15.381795Z","status":"completed","version":"1.0.0","findings_count":0}],"snapshot_sha256":"b9d264c700af20b793f1d7bf1693dcbba2479fb34bdfac9841d0021795af17f9"},"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"}