{"paper":{"title":"Quantifying Potential Observation Missingness in Inverse Reinforcement Learning","license":"http://creativecommons.org/licenses/by/4.0/","headline":"Missing observations in IRL can be quantified by finding the minimal perturbations that make expert actions appear optimal.","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Abhishek Sharma, Alihan Huyuk, Finale Doshi-Velez, Leo Benac","submitted_at":"2026-05-12T23:56:36Z","abstract_excerpt":"Inverse reinforcement learning (IRL), which infers reward functions from demonstrations, is a valuable tool for modeling and understanding decision-making behavior. Many variants of IRL have been developed to capture complexities of human decision-making, such as subjective beliefs, imperfect planning, and dynamic goals. However, an often-overlooked issue in real-world behavioral datasets is that the recorded data may be missing observations that were available to the original decision-maker. In use-inspired settings such as healthcare, this can make expert actions appear suboptimal, even when"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"We identify the minimal perturbations to the recorded observations needed for the expert's actions to appear optimal. We develop a practical algorithm for this problem and demonstrate its utility for quantifying the possible extent of missing observations in behavioral datasets through extensive experiments on synthetic navigation tasks, a cancer treatment simulator, and ICU treatment data.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That the minimal perturbations identified correspond to plausible unobserved observations available to the original decision-maker and that standard IRL optimality assumptions hold once those observations are restored.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"A practical algorithm quantifies potential missing observations in IRL by computing minimal perturbations to recorded data that render expert actions optimal.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"Missing observations in IRL can be quantified by finding the minimal perturbations that make expert actions appear optimal.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"00779c056a8109a15852afebe4b6be1866a9d8f49a6d01690855a33d2c290674"},"source":{"id":"2605.12831","kind":"arxiv","version":1},"verdict":{"id":"97210000-b0a9-447e-9b1c-5d12342cadd5","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-14T19:40:05.127295Z","strongest_claim":"We identify the minimal perturbations to the recorded observations needed for the expert's actions to appear optimal. We develop a practical algorithm for this problem and demonstrate its utility for quantifying the possible extent of missing observations in behavioral datasets through extensive experiments on synthetic navigation tasks, a cancer treatment simulator, and ICU treatment data.","one_line_summary":"A practical algorithm quantifies potential missing observations in IRL by computing minimal perturbations to recorded data that render expert actions optimal.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That the minimal perturbations identified correspond to plausible unobserved observations available to the original decision-maker and that standard IRL optimality assumptions hold once those observations are restored.","pith_extraction_headline":"Missing observations in IRL can be quantified by finding the minimal perturbations that make expert actions appear optimal."},"references":{"count":43,"sample":[{"doi":"","year":null,"title":"Algorithms for inverse reinforcement learning. , author=. International Conference on Machine Learning , year=","work_id":"f5d97198-9e72-435a-b8e4-1698fcfed900","ref_index":1,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":null,"title":"Bayesian Inverse Transition Learning: Learning Dynamics From Near-Optimal Trajectories","work_id":"5511fe45-1a49-4f94-ad48-f4ec610da3ee","ref_index":2,"cited_arxiv_id":"2411.05174","is_internal_anchor":true},{"doi":"","year":2013,"title":"New England Journal of Medicine , volume=","work_id":"0a66328c-dd13-48b8-896c-2c0333b9aff4","ref_index":3,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2017,"title":"Canadian Journal of Anesthesia/Journal canadien d'anesth","work_id":"0b356b15-cfaa-409f-aa84-67fb5e89a0f4","ref_index":4,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2022,"title":"Journal of Critical Care , volume=","work_id":"4d77bf63-09d4-4add-82af-4dfd49962af3","ref_index":5,"cited_arxiv_id":"","is_internal_anchor":false}],"resolved_work":43,"snapshot_sha256":"59e29770ab8e228eedc25732d766f84a4fb8c51fe87fe6b90efb0d569683c329","internal_anchors":3},"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"}