{"paper":{"title":"TILT: Target-induced loss tilting under covariate shift","license":"http://creativecommons.org/licenses/by/4.0/","headline":"The target-side penalty on an auxiliary predictor component induces implicit relative importance weighting that stays bounded even with disjoint supports.","cross_cats":["stat.ML"],"primary_cat":"cs.LG","authors_text":"Kakei Yamamoto, Martin J. Wainwright","submitted_at":"2026-05-14T02:26:34Z","abstract_excerpt":"We introduce and analyze Target-Induced Loss Tilting (TILT) for unsupervised domain adaptation under covariate shift. It is based on a novel objective function that decomposes the source predictor as $f+b$, fits $f+b$ on labeled source data while simultaneously penalizing the auxiliary component $b$ on unlabeled target inputs. The resulting fit $f$ is deployed as the final target predictor. At the population level, we show that this target-side penalty implicitly induces relative importance weighting at the population level, but in terms of an estimand $b^*_f$ that is self-localized to the cur"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"At the population level, the target-side penalty on b implicitly induces relative importance weighting in terms of an estimand b*_f that is self-localized to the current error and remains uniformly bounded for any source-target pair, even those with disjoint supports; a general finite-sample oracle inequality holds and yields an end-to-end guarantee for sparse ReLU networks.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"The analysis assumes the existence of a decomposition f + b where the penalty on b on target data produces a useful weighting without requiring the supports of source and target to overlap or any explicit density estimation.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"TILT adds a target-data penalty on an auxiliary predictor component to induce effective importance weighting for unsupervised domain adaptation under covariate shift.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"The target-side penalty on an auxiliary predictor component induces implicit relative importance weighting that stays bounded even with disjoint supports.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"127adcba1b860c889fda8a3d77108e87cd85f7c456209098ac185230b1237569"},"source":{"id":"2605.14280","kind":"arxiv","version":1},"verdict":{"id":"d770991c-ad92-4e37-ad65-f797ca48a70a","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-15T02:25:24.880973Z","strongest_claim":"At the population level, the target-side penalty on b implicitly induces relative importance weighting in terms of an estimand b*_f that is self-localized to the current error and remains uniformly bounded for any source-target pair, even those with disjoint supports; a general finite-sample oracle inequality holds and yields an end-to-end guarantee for sparse ReLU networks.","one_line_summary":"TILT adds a target-data penalty on an auxiliary predictor component to induce effective importance weighting for unsupervised domain adaptation under covariate shift.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"The analysis assumes the existence of a decomposition f + b where the penalty on b on target data produces a useful weighting without requiring the supports of source and target to overlap or any explicit density estimation.","pith_extraction_headline":"The target-side penalty on an auxiliary predictor component induces implicit relative importance weighting that stays bounded even with disjoint supports."},"references":{"count":144,"sample":[{"doi":"","year":null,"title":"Tyrrell Rockafellar and Ren","work_id":"ba3a8e18-5c32-4b7e-86c0-775112546ed9","ref_index":1,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2009,"title":"2009 , publisher =","work_id":"21422820-909c-4f56-9bfd-76bc6834a77d","ref_index":2,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":null,"title":"Tyrrell Rockafellar and Roger J.-B","work_id":"923f93d7-5e68-4db1-9247-7c11527b2706","ref_index":3,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2014,"title":"Foundations and Trends in Optimization , volume =","work_id":"594ac684-7262-42d4-8272-bad3725b35d3","ref_index":4,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2016,"title":"Proceedings of the 29th Annual Conference on Learning Theory , series =","work_id":"8ec44617-4c2f-4e9f-9978-18b58ca564eb","ref_index":5,"cited_arxiv_id":"","is_internal_anchor":false}],"resolved_work":144,"snapshot_sha256":"7ad11daebb02e721965cbfd6ccc29c3ff70d93ca8b4cc5a8b65013050517275a","internal_anchors":4},"formal_canon":{"evidence_count":2,"snapshot_sha256":"31ce2be4f2c59a0461c92f596d01663f3dfaddf4aa7204d1349bf7d0cdf8eefc"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"}