{"paper":{"title":"Aligning Network Equivariance with Data Symmetry: A Theoretical Framework and Adaptive Approach for Image Restoration","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"Equivariance error of the optimal restoration operator is strictly bounded by data symmetry error and discretization mesh size.","cross_cats":[],"primary_cat":"cs.CV","authors_text":"Deyu Meng, Feiyu Tan, Qi Xie, Zongben Xu","submitted_at":"2026-05-13T16:22:19Z","abstract_excerpt":"Image restoration is an inherently ill posed inverse problem. Equivariant networks that embed geometric symmetry priors can mitigate this ill posedness and improve performance. However, current understanding of the relationship between network equivariance and data symmetry remains largely heuristic. Particularly for real world data with imperfect symmetry, existing research lacks a systematic theoretical framework to quantify symmetry, select transformation groups, or evaluate model data alignment. To bridge this gap, we conduct an analysis from an optimization perspective and formalize the i"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"the equivariance error of the optimal restoration operator is strictly bounded by the data symmetry error and the discretization mesh size. Furthermore, by analyzing the network's empirical risk, we demonstrate that aligning equivariance with data symmetry optimizes the bias variance trade off, minimizing the total expected risk.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"that the proposed quantifiable definition of non-strict symmetry at the dataset level (rather than sample level) can be used as a valid constraint to formulate the restoration inverse problem for real-world data with imperfect symmetry.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"A new dataset-level non-strict symmetry measure allows deriving bounded equivariance for restoration models and motivates an adaptive network that aligns with per-sample symmetry to reduce expected risk.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"Equivariance error of the optimal restoration operator is strictly bounded by data symmetry error and discretization mesh size.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"7860202c945dceeb2fff636bb6a566b799039b63c77ec856c0e8852079442500"},"source":{"id":"2605.13744","kind":"arxiv","version":1},"verdict":{"id":"196eb336-d3cb-4e32-9a20-a0e6a16b069f","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-14T19:48:24.213823Z","strongest_claim":"the equivariance error of the optimal restoration operator is strictly bounded by the data symmetry error and the discretization mesh size. Furthermore, by analyzing the network's empirical risk, we demonstrate that aligning equivariance with data symmetry optimizes the bias variance trade off, minimizing the total expected risk.","one_line_summary":"A new dataset-level non-strict symmetry measure allows deriving bounded equivariance for restoration models and motivates an adaptive network that aligns with per-sample symmetry to reduce expected risk.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"that the proposed quantifiable definition of non-strict symmetry at the dataset level (rather than sample level) can be used as a valid constraint to formulate the restoration inverse problem for real-world data with imperfect symmetry.","pith_extraction_headline":"Equivariance error of the optimal restoration operator is strictly bounded by data symmetry error and discretization mesh size."},"references":{"count":70,"sample":[{"doi":"","year":1991,"title":"Communications of the ACM , volume=","work_id":"f6ba041d-76f6-48ab-9f0a-f8e95e68ae5c","ref_index":1,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2019,"title":"IEEE transactions on pattern analysis and machine intelligence , volume=","work_id":"a8d4203c-ef45-4f3f-96f8-e7a17ee8197d","ref_index":2,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2002,"title":"IEEE Transactions on Acoustics, Speech, and Signal Processing , volume=","work_id":"6818ca35-23ff-4816-af74-8b6121664bba","ref_index":3,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2008,"title":"SIAM Journal on Scientific Computing , volume=","work_id":"ad332d81-cce1-4ec4-84f5-8b3caa7d3139","ref_index":4,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2021,"title":"IEEE Geoscience and Remote Sensing Magazine , volume=","work_id":"e49b17a2-0eac-484e-bb4c-aeb923761a10","ref_index":5,"cited_arxiv_id":"","is_internal_anchor":false}],"resolved_work":70,"snapshot_sha256":"e0c555db00477ff2ce0184ee1e7c951db4f191921b0be34c3450ac5d1e11a1cf","internal_anchors":2},"formal_canon":{"evidence_count":2,"snapshot_sha256":"54818977e59dede24c5d1c732cf62539975a28b90f61f860cd447218289b9d11"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"}