{"paper":{"title":"What Information Matters? Graph Out-of-Distribution Detection via Tri-Component Information Decomposition","license":"http://creativecommons.org/licenses/by/4.0/","headline":"TIDE decomposes graph information into feature-specific, structure-specific and joint components to retain only label-relevant joint signals for improved out-of-distribution node detection.","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Danny Wang, Ruihong Qiu, Zi Huang","submitted_at":"2026-05-13T05:36:33Z","abstract_excerpt":"Graph neural networks are widely used for node classification, but they remain vulnerable to out-of-distribution (OOD) shifts in node features and graph structure. Prior work established that methods trained with standard supervised learning (SL) objectives tend to capture spurious signals from either features and/or structure, leaving the model fragile under distributional changes. To address this, we propose TIDE, a novel and effective Tri-Component Information Decomposition framework that explicitly decomposes information into feature-specific, structure-specific and joint components. TIDE "},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"TIDE explicitly decomposes information into feature-specific, structure-specific and joint components, preserving only the label-relevant part of the joint information while filtering out spurious feature- and structure-specific information.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That the joint information component can be reliably separated into label-relevant versus spurious parts and that removing the spurious specific components will produce a measurable entropy gap and higher ID confidence without harming ID accuracy.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"TIDE decomposes graph information into feature-specific, structure-specific, and joint components to retain only label-relevant joint signals and improve OOD detection over standard supervised learning.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"TIDE decomposes graph information into feature-specific, structure-specific and joint components to retain only label-relevant joint signals for improved out-of-distribution node detection.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"8e7c9f5280723e3d6f52cbda363786486aba2ee10cbcd2aee818bbb5309f9d39"},"source":{"id":"2605.13032","kind":"arxiv","version":2},"verdict":{"id":"e21fc5ea-96cf-45fb-9db4-049bdf6edc16","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-15T05:37:48.342808Z","strongest_claim":"TIDE explicitly decomposes information into feature-specific, structure-specific and joint components, preserving only the label-relevant part of the joint information while filtering out spurious feature- and structure-specific information.","one_line_summary":"TIDE decomposes graph information into feature-specific, structure-specific, and joint components to retain only label-relevant joint signals and improve OOD detection over standard supervised learning.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That the joint information component can be reliably separated into label-relevant versus spurious parts and that removing the spurious specific components will produce a measurable entropy gap and higher ID confidence without harming ID accuracy.","pith_extraction_headline":"TIDE decomposes graph information into feature-specific, structure-specific and joint components to retain only label-relevant joint signals for improved out-of-distribution node detection."},"references":{"count":133,"sample":[{"doi":"","year":null,"title":"Your classifier is secretly an energy based model and you should treat it like one , booktitle =","work_id":"02851fc1-372f-4f7b-b257-bf4a51cb34b6","ref_index":1,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":null,"title":"Srikant , title =","work_id":"69a90bef-a8a1-45db-bb42-08718f01383e","ref_index":2,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":null,"title":"CVPR , year =","work_id":"62d26c3e-fe18-48f7-a13a-b33c009ae852","ref_index":3,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":null,"title":"NeurIPS , year =","work_id":"88297c32-cee8-4ca9-a5eb-71b2c2b3160b","ref_index":4,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":null,"title":"NeurIPS , year =","work_id":"b2dbd2af-dfa7-4f2f-8c46-c69bc1c25f68","ref_index":5,"cited_arxiv_id":"","is_internal_anchor":false}],"resolved_work":133,"snapshot_sha256":"1278d4eb508c175fc77f279873cd29b383e4d2479182613ee45e92f2d6ba2e70","internal_anchors":0},"formal_canon":{"evidence_count":2,"snapshot_sha256":"3cd33cf55920df7773abb0808e8dabac86fcd6181f24c597bcdf4f3b1df90544"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"}