{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2026:L66OCX2VMFITANOO7VUXRY7J76","merge_version":"pith-open-graph-merge-v1","event_count":2,"valid_event_count":2,"invalid_event_count":0,"equivocation_count":0,"current":{"canonical_record":{"metadata":{"abstract_canon_sha256":"feb1200c667e62499a964fa0350b9855e7bae6f42b91582fecb2e441b879e76a","cross_cats_sorted":["cs.CV"],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2026-05-13T11:51:46Z","title_canon_sha256":"81bea3319b199b02d255251cd202e5accab761dfd22b0d4d2c4bb7980e3ac8a9"},"schema_version":"1.0","source":{"id":"2605.13395","kind":"arxiv","version":1}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2605.13395","created_at":"2026-05-18T02:44:47Z"},{"alias_kind":"arxiv_version","alias_value":"2605.13395v1","created_at":"2026-05-18T02:44:47Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.13395","created_at":"2026-05-18T02:44:47Z"},{"alias_kind":"pith_short_12","alias_value":"L66OCX2VMFIT","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_16","alias_value":"L66OCX2VMFITANOO","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_8","alias_value":"L66OCX2V","created_at":"2026-05-18T12:33:37Z"}],"graph_snapshots":[{"event_id":"sha256:66cd2496b5d3f5358543721e8298a75bac193aab4121380ce4faf81bfd8cd5cb","target":"graph","created_at":"2026-05-18T02:44:47Z","signer":{"key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signer_id":"pith.science","signer_type":"pith_registry"},"payload":{"graph_snapshot":{"author_claims":{"count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","strong_count":0},"builder_version":"pith-number-builder-2026-05-17-v1","claims":{"count":4,"items":[{"attestation":"unclaimed","claim_id":"C1","kind":"strongest_claim","source":"verdict.strongest_claim","status":"machine_extracted","text":"perturbations can simultaneously address both adversarial vulnerability and class imbalance. Based on these insights, we propose RobustLT, a plug-and-play framework that adaptively adjusts perturbations during adversarial training. Extensive experiments demonstrate that RobustLT consistently enhances adversarial robustness and class-balance on long-tailed datasets."},{"attestation":"unclaimed","claim_id":"C2","kind":"weakest_assumption","source":"verdict.weakest_assumption","status":"machine_extracted","text":"The theoretical claim that perturbations inherently alter the training distribution in a way that simultaneously fixes both skew and instability holds without introducing new instabilities or requiring dataset-specific tuning beyond the adaptive rule."},{"attestation":"unclaimed","claim_id":"C3","kind":"one_line_summary","source":"verdict.one_line_summary","status":"machine_extracted","text":"RobustLT adaptively adjusts perturbations in adversarial training to simultaneously improve robustness and class balance on long-tailed datasets."},{"attestation":"unclaimed","claim_id":"C4","kind":"headline","source":"verdict.pith_extraction.headline","status":"machine_extracted","text":"Adaptive perturbations can rebalance class distributions during adversarial training"}],"snapshot_sha256":"31b25fc73290412ca1f82700397f79dd0f5680f4178fc6a38fbd0b2eee16d3d5"},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"paper":{"abstract_excerpt":"Deep neural networks are highly vulnerable to adversarial examples, i.e.,small perturbations that can significantly degrade model performance. While adversarial training has become the primary defense strategy, most studies focus on balanced datasets, overlooking the challenges posed by real-world long-tail data. Motivated by the fact that perturbations in adversarial examples inherently alter the training distribution, we theoretically investigate their impact. We first revisit adversarial training for long-tail data and identify two key limitations: (i) a skewed training objective caused by ","authors_text":"Jiancheng Shi, Lilin Zhang, Xianggen Liu, Yimo Guo, Yue Li","cross_cats":["cs.CV"],"headline":"Adaptive perturbations can rebalance class distributions during adversarial training","license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2026-05-13T11:51:46Z","title":"Taming the Long Tail: Rebalancing Adversarial Training via Adaptive Perturbation"},"references":{"count":53,"internal_anchors":2,"resolved_work":53,"sample":[{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":1,"title":"Recent advances in adversarial training for adversarial ro- bustness","work_id":"7a75d48d-725e-4dce-b414-42bd408a421c","year":2021},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":2,"title":"Evasion attacks against machine learning at test time","work_id":"ac65a4a1-63d7-49ce-a47d-5aff7fa40b10","year":2013},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":3,"title":"A systematic study of the class imbalance problem in convolu- tional neural networks.Neural Networks, 106:249–259, 2018","work_id":"dc797dde-e844-4c97-a15f-baec9aadb7a0","year":2018},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":4,"title":"A unified wasserstein distributional robustness framework for adversarial training","work_id":"b3fe9fa4-480f-44eb-9f95-f6ea0d345127","year":2022},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":5,"title":"Learning imbalanced datasets with label- distribution-aware margin loss.Advances in Neural Informa- tion Processing Systems, 32, 2019","work_id":"09cd6780-357e-4735-bd39-a777abc8db35","year":2019}],"snapshot_sha256":"86c84585ea604857b282401dfa4aedf97ff7342b73e7b1c6de160be9e894ba5c"},"source":{"id":"2605.13395","kind":"arxiv","version":1},"verdict":{"created_at":"2026-05-14T20:15:49.398078Z","id":"4ed64c83-c395-4ae8-9355-f0a2afa2d68c","model_set":{"reader":"grok-4.3"},"one_line_summary":"RobustLT adaptively adjusts perturbations in adversarial training to simultaneously improve robustness and class balance on long-tailed datasets.","pipeline_version":"pith-pipeline@v0.9.0","pith_extraction_headline":"Adaptive perturbations can rebalance class distributions during adversarial training","strongest_claim":"perturbations can simultaneously address both adversarial vulnerability and class imbalance. Based on these insights, we propose RobustLT, a plug-and-play framework that adaptively adjusts perturbations during adversarial training. Extensive experiments demonstrate that RobustLT consistently enhances adversarial robustness and class-balance on long-tailed datasets.","weakest_assumption":"The theoretical claim that perturbations inherently alter the training distribution in a way that simultaneously fixes both skew and instability holds without introducing new instabilities or requiring dataset-specific tuning beyond the adaptive rule."}},"verdict_id":"4ed64c83-c395-4ae8-9355-f0a2afa2d68c"}}],"author_attestations":[],"timestamp_anchors":[],"storage_attestations":[],"citation_signatures":[],"replication_records":[],"corrections":[],"mirror_hints":[],"record_created":{"event_id":"sha256:3259edd2d93e0c6a5cbd678b441e41443bac2d091cb401ac25f8f927909a7d1a","target":"record","created_at":"2026-05-18T02:44:47Z","signer":{"key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signer_id":"pith.science","signer_type":"pith_registry"},"payload":{"attestation_state":"computed","canonical_record":{"metadata":{"abstract_canon_sha256":"feb1200c667e62499a964fa0350b9855e7bae6f42b91582fecb2e441b879e76a","cross_cats_sorted":["cs.CV"],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2026-05-13T11:51:46Z","title_canon_sha256":"81bea3319b199b02d255251cd202e5accab761dfd22b0d4d2c4bb7980e3ac8a9"},"schema_version":"1.0","source":{"id":"2605.13395","kind":"arxiv","version":1}},"canonical_sha256":"5fbce15f5561513035cefd6978e3e9ffb81660fc3fe2b964af37e5836f5ff2b6","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"5fbce15f5561513035cefd6978e3e9ffb81660fc3fe2b964af37e5836f5ff2b6","first_computed_at":"2026-05-18T02:44:47.666009Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-05-18T02:44:47.666009Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"C576gbbhzqXXN45HG3baPxMnq5cqrEbIlLxFxXKXGkaECWKUbNdh3cATsP2ZIG9fmY9jEqxOG7poa5O3xbbkAg==","signature_status":"signed_v1","signed_at":"2026-05-18T02:44:47.666461Z","signed_message":"canonical_sha256_bytes"},"source_id":"2605.13395","source_kind":"arxiv","source_version":1}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:3259edd2d93e0c6a5cbd678b441e41443bac2d091cb401ac25f8f927909a7d1a","sha256:66cd2496b5d3f5358543721e8298a75bac193aab4121380ce4faf81bfd8cd5cb"],"state_sha256":"4f82824de835f7afdb4a53e708e5dafa8f743ab4fe90d71caeaba4ee0eaad60d"}