{
  "article_slug": "mixup-vicinal-risk",
  "claims": [
    {
      "id": "s1",
      "kind": "author_analysis_or_verification_proposal",
      "sources": [],
      "claim": "1. Interpolation is a statistical intervention"
    },
    {
      "id": "s2",
      "kind": "source_based_review",
      "sources": [
        "dl3-b-mixup-vicinal-risk-1",
        "dl3-b-mixup-vicinal-risk-2",
        "dl3-b-mixup-vicinal-risk-3",
        "dl3-b-mixup-vicinal-risk-4",
        "dl3-b-mixup-vicinal-risk-5",
        "dl3-b-mixup-vicinal-risk-6"
      ],
      "claim": "2. The primary literature supports several mechanisms"
    },
    {
      "id": "s3",
      "kind": "author_analysis_or_verification_proposal",
      "sources": [],
      "claim": "3. Define the vicinal distribution explicitly"
    },
    {
      "id": "s4",
      "kind": "author_analysis_or_verification_proposal",
      "sources": [],
      "claim": "4. Two endpoints do not determine the correct interior"
    },
    {
      "id": "s5",
      "kind": "author_analysis_or_verification_proposal",
      "sources": [],
      "claim": "5. Classification accuracy can conceal probability mismatch"
    },
    {
      "id": "s6",
      "kind": "author_analysis_or_verification_proposal",
      "sources": [],
      "claim": "6. Mixing strength is not summarized by the mean weight"
    },
    {
      "id": "s7",
      "kind": "author_analysis_or_verification_proposal",
      "sources": [],
      "claim": "7. A quadratic expansion reveals an added constraint"
    },
    {
      "id": "s8",
      "kind": "author_analysis_or_verification_proposal",
      "sources": [],
      "claim": "8. Input geometry and semantic geometry can disagree"
    },
    {
      "id": "s9",
      "kind": "author_analysis_or_verification_proposal",
      "sources": [],
      "claim": "9. Hidden-state interpolation inherits a learned coordinate system"
    },
    {
      "id": "s10",
      "kind": "author_analysis_or_verification_proposal",
      "sources": [],
      "claim": "10. Spatial mixing uses a different label assumption"
    },
    {
      "id": "s11",
      "kind": "author_analysis_or_verification_proposal",
      "sources": [],
      "claim": "11. Calibration must be evaluated on the target distribution"
    },
    {
      "id": "s12",
      "kind": "author_analysis_or_verification_proposal",
      "sources": [],
      "claim": "12. Robustness claims need a specified perturbation set"
    },
    {
      "id": "s13",
      "kind": "author_analysis_or_verification_proposal",
      "sources": [],
      "claim": "13. Pairing policies can change class and source balance"
    },
    {
      "id": "s14",
      "kind": "author_analysis_or_verification_proposal",
      "sources": [],
      "claim": "14. A discriminating experimental design"
    },
    {
      "id": "s15",
      "kind": "author_analysis_or_verification_proposal",
      "sources": [],
      "claim": "15. Executable checks establish the limited analytical claims"
    },
    {
      "id": "s16",
      "kind": "author_analysis_or_verification_proposal",
      "sources": [],
      "claim": "16. Implications for interpretable augmentation research"
    }
  ]
}