{"paper":{"title":"IGT-OMD: Implicit Gradient Transport for Decision-Focused Learning under Delayed Feedback","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"IGT-OMD corrects gradient staleness in delayed bilevel optimization by re-evaluating stale gradients at current parameters using stored inner solutions.","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Benjamin Amoh, Geoffrey G. Parker, Wesley Marrero","submitted_at":"2026-05-12T19:43:49Z","abstract_excerpt":"Decision-focused learning trains predictive models end-to-end against downstream decision loss, but online settings suffer delayed feedback: outcomes may not arrive for many environment interactions. We identify \\emph{staleness amplification}, a failure mode unique to bilevel optimization under delay, in which gradient staleness couples with inner-solver sensitivity to inflate regret beyond single-level delay theory. We prove that any black-box delayed optimizer incurs an irreducible regret cost from inner-solver approximation error, and that gradient staleness contributes a quadratically grow"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"IGT-OMD achieves the first sublinear regret bound for delayed bilevel optimization with queue-length-adaptive step sizes by reducing transport error from quadratic to linear dependence on delay.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That inner solutions can be stored and re-evaluated at current parameters with negligible extra cost and that the bilevel problem satisfies the smoothness and convexity conditions needed for the regret analysis to go through.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"IGT-OMD reduces gradient transport error from quadratic to linear in delay length for delayed bilevel optimization and achieves sublinear regret with adaptive steps.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"IGT-OMD corrects gradient staleness in delayed bilevel optimization by re-evaluating stale gradients at current parameters using stored inner solutions.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"3d27e44032881670387e6a6ac7fa6fc7976efda9282217cd3c46d6274baa01cb"},"source":{"id":"2605.12693","kind":"arxiv","version":1},"verdict":{"id":"8f86830b-418b-41bc-a2c4-eae6fe4b7653","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-14T21:00:32.254926Z","strongest_claim":"IGT-OMD achieves the first sublinear regret bound for delayed bilevel optimization with queue-length-adaptive step sizes by reducing transport error from quadratic to linear dependence on delay.","one_line_summary":"IGT-OMD reduces gradient transport error from quadratic to linear in delay length for delayed bilevel optimization and achieves sublinear regret with adaptive steps.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That inner solutions can be stored and re-evaluated at current parameters with negligible extra cost and that the bilevel problem satisfies the smoothness and convexity conditions needed for the regret analysis to go through.","pith_extraction_headline":"IGT-OMD corrects gradient staleness in delayed bilevel optimization by re-evaluating stale gradients at current parameters using stored inner solutions."},"references":{"count":43,"sample":[{"doi":"10.1287/mnsc.2020.3922","year":2022,"title":"Adam N. Elmachtoub and Paul Grigas. Smart “Predict, then Optimize”.Management Science, 68(1):9–26, 2022. doi: 10.1287/mnsc.2020.3922","work_id":"1d367d07-9caa-459c-a28c-8c2f98a58c01","ref_index":1,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"10.1609/aaai.v33i01","year":2019,"title":"Melding the data-decisions pipeline: Decision- focused learning for combinatorial optimization","work_id":"15e6042d-4da3-4672-a28c-8bb3a3ce644a","ref_index":2,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2024,"title":"Decision-focused learning: Foundations, state of the art, benchmark and future opportunities.Journal of Artificial Intelligence Research, 81:1623–1701, 2024","work_id":"e88758ea-7402-4199-a2cd-3acb11457338","ref_index":3,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2013,"title":"Online learning under delayed feedback","work_id":"61dd1aaa-43c4-4c48-a334-5b085d37244c","ref_index":4,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2012,"title":"Online learning and online convex optimization.Foundations and Trends in Machine Learning, 4(2):107–194","work_id":"3bd3c6eb-fd04-4237-95c3-1ada8bc8a1ae","ref_index":5,"cited_arxiv_id":"","is_internal_anchor":false}],"resolved_work":43,"snapshot_sha256":"3f0569b265d25d679f0baa4762bf968c9108f42a61e9ca78cee3cc70962dcf03","internal_anchors":4},"formal_canon":{"evidence_count":2,"snapshot_sha256":"0dd4b6dd7147c262c070828b81601954cdbbdead2e53649b1ade512add586e18"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"}