{"paper":{"title":"U-CECE: A Universal Multi-Resolution Framework for Conceptual Counterfactual Explanations","license":"http://creativecommons.org/licenses/by-nc-nd/4.0/","headline":"U-CECE generates conceptual counterfactual explanations at three detail levels with neural approximations that match or exceed exact graph methods in human preference.","cross_cats":["cs.CV"],"primary_cat":"cs.AI","authors_text":"Angeliki Dimitriou, Giorgos Filandrianos, Giorgos Stamou, Maria Lymperaiou, Nikolaos Chaidos","submitted_at":"2026-04-09T14:30:00Z","abstract_excerpt":"As AI models grow more complex, explainability is essential for building trust, yet concept-based counterfactual methods still face a trade-off between expressivity and efficiency. Representing underlying concepts as atomic sets is fast but misses relational context, whereas full graph representations are more faithful but require solving the NP-hard Graph Edit Distance (GED) problem. We propose U-CECE, a unified, model-agnostic multi-resolution framework for conceptual counterfactual explanations that adapts to data regime and compute budget. U-CECE spans three levels of expressivity: atomic "},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"human surveys and LVLM-based evaluation show that the retrieved structural counterfactuals are semantically equivalent to, and often preferred over, exact GED-based ground-truth explanations.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That the chosen graph representations and the transductive GNN / inductive GAE approximations faithfully encode the semantic relations required for valid conceptual counterfactuals without introducing systematic distortions.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"U-CECE offers a unified model-agnostic framework spanning atomic, relational, and structural levels for conceptual counterfactuals, with structural GNN/GAE modes producing explanations semantically equivalent or preferred to exact GED ground truth on CUB and Visual Genome.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"U-CECE generates conceptual counterfactual explanations at three detail levels with neural approximations that match or exceed exact graph methods in human preference.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"b53abcd11f9bafd88b0d1ed559a22877dfcc2f242d0b25dcf72233ac902aa3f3"},"source":{"id":"2604.08295","kind":"arxiv","version":3},"verdict":{"id":"5cd24ab9-1201-4730-ae7c-8d14f83e06f3","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-10T18:19:46.387872Z","strongest_claim":"human surveys and LVLM-based evaluation show that the retrieved structural counterfactuals are semantically equivalent to, and often preferred over, exact GED-based ground-truth explanations.","one_line_summary":"U-CECE offers a unified model-agnostic framework spanning atomic, relational, and structural levels for conceptual counterfactuals, with structural GNN/GAE modes producing explanations semantically equivalent or preferred to exact GED ground truth on CUB and Visual Genome.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That the chosen graph representations and the transductive GNN / inductive GAE approximations faithfully encode the semantic relations required for valid conceptual counterfactuals without introducing systematic distortions.","pith_extraction_headline":"U-CECE generates conceptual counterfactual explanations at three detail levels with neural approximations that match or exceed exact graph methods in human preference."},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2604.08295/integrity.json","findings":[],"available":true,"detectors_run":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938"},"references":{"count":0,"sample":[],"resolved_work":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","internal_anchors":0},"formal_canon":{"evidence_count":2,"snapshot_sha256":"63f02b7a32d8f4fbef2739f58d6dfdb7d0301d767848e75052b437e9b83acd81"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"}