{"paper":{"title":"DRIFT: A Benchmark for Task-Free Continual Graph Learning with Continuous Distribution Shifts","license":"http://creativecommons.org/licenses/by/4.0/","headline":"Many existing continual graph learning methods implicitly depend on task boundaries and degrade under continuous distribution shifts in task-free streams.","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Dongjin Song, Guiquan Sun, Jingchao Ni, Xikun Zhang","submitted_at":"2026-05-13T04:54:46Z","abstract_excerpt":"Continual graph learning (CGL) aims to learn from dynamically evolving graphs while mitigating catastrophic forgetting. Existing CGL approaches typically adopt a task-based formulation, where the data stream is partitioned into a sequence of discrete tasks with pre-defined boundaries. However, such assumptions rarely hold in real-world environments, where data distributions evolve continuously and task identity is often unavailable. To better reflect realistic non-stationary environments, we revisit continual graph learning from a task-free perspective. We propose a unified formulation that mo"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"Our findings indicate that many existing approaches implicitly rely on task boundary information and struggle under realistic task-free graph streams.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"The Gaussian parameterization of transition dynamics between latent task distributions accurately captures the spectrum of real-world continuous distribution shifts in graph data.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"DRIFT benchmark shows substantial performance degradation for continual graph learning methods under task-free continuous distribution shifts modeled via Gaussian mixtures.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"Many existing continual graph learning methods implicitly depend on task boundaries and degrade under continuous distribution shifts in task-free streams.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"b7e567077d7062c3d3d931e9e87470aa1258a55f80f6ccb8f9283abc454d2da7"},"source":{"id":"2605.12998","kind":"arxiv","version":2},"verdict":{"id":"086289f7-17a4-4c91-a067-95e44fb45f61","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-15T05:42:58.834938Z","strongest_claim":"Our findings indicate that many existing approaches implicitly rely on task boundary information and struggle under realistic task-free graph streams.","one_line_summary":"DRIFT benchmark shows substantial performance degradation for continual graph learning methods under task-free continuous distribution shifts modeled via Gaussian mixtures.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"The Gaussian parameterization of transition dynamics between latent task distributions accurately captures the spectrum of real-world continuous distribution shifts in graph data.","pith_extraction_headline":"Many existing continual graph learning methods implicitly depend on task boundaries and degrade under continuous distribution shifts in task-free streams."},"references":{"count":51,"sample":[{"doi":"","year":2017,"title":"Inductive representation learning on large graphs","work_id":"ee217072-19c9-4225-b00c-43701e678a27","ref_index":1,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2020,"title":"Open graph benchmark: Datasets for machine learning on graphs","work_id":"39cc4a5f-a4c0-4922-a15f-e07bc2f727e8","ref_index":2,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2017,"title":"Knowledge graph embedding: A survey of approaches and applications.IEEE transactions on knowledge and data engineering, 29:2724– 2743, 2017","work_id":"f7234f26-5339-4a43-866a-dc905f24ca7a","ref_index":3,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2017,"title":"Overcoming catastrophic forgetting in neural networks.Proceedings of the national academy of sciences, 114(13):3521–3526","work_id":"68b7964f-cd25-47e6-a060-52bbc86a995d","ref_index":4,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":1902,"title":"Chaudhry et al","work_id":"001da84d-4329-4699-85e9-5431d143af9b","ref_index":5,"cited_arxiv_id":"1902.10486","is_internal_anchor":true}],"resolved_work":51,"snapshot_sha256":"b27d8fa8dbd02e8bdd2bfecf74b652ce5d94dd97daba7999229cbed6fac04de3","internal_anchors":4},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"}