{
  "schema_version": 1,
  "title": "Repro: Coupled Training with Privileged Information and Unlabeled Data",
  "emoji": "microscope",
  "space_id": "snaykey/repro-coupled-privileged-training",
  "paper": {
    "openreview_id": "WlCYSqjnqS"
  },
  "tags": [
    "icml2026-repro",
    "paper-WlCYSqjnqS"
  ],
  "updated_at": "2026-07-27T00:00:00+00:00",
  "root": {
    "slug": "index",
    "title": "Repro: Coupled Training with Privileged Information and Unlabeled Data",
    "file": "pages/index.md",
    "children": [
      {
        "slug": "executive-summary",
        "title": "Executive summary",
        "file": "pages/executive-summary/page.md",
        "children": []
      },
      {
        "slug": "claim-1",
        "title": "The coupled training framework jointly optimizes a deployment model using only test-time features X and a rich-view model using both X and privileged features W, with population minimizers characterized via fixed-point equations interpolating between the deployment target μ(x) and rich-view regression function η(z) (Theorem 2.1).",
        "file": "pages/claim-1/page.md",
        "children": []
      },
      {
        "slug": "claim-2",
        "title": "The risk bound depends on a correlation coefficient ρ* measuring alignment between deployment and rich-view estimation errors, and the method provably never performs worse than supervised learning on labeled data alone when ρ* is large (Corollary 2.3, Theorem 2.7).",
        "file": "pages/claim-2/page.md",
        "children": []
      },
      {
        "slug": "claim-3",
        "title": "An alternating forward selection procedure for the high-dimensional extension is proven to achieve sublinear decay of the empirical objective with O(N(|D_f|+|D_g|)k) time complexity (Theorem 3.1).",
        "file": "pages/claim-3/page.md",
        "children": []
      },
      {
        "slug": "claim-4",
        "title": "On Parkinson's telemonitoring regression, coupled training achieves test MSE of 72.60 versus 77.34 for the baseline using voice measurements with spectral features as privileged information (Table 1).",
        "file": "pages/claim-4/page.md",
        "children": []
      },
      {
        "slug": "claim-5",
        "title": "On Bank Marketing classification, coupled training attains mean Brier score 0.0881, outperforming Baseline (0.0893), Two-Stage pseudo-labeling (0.0928), and SVM+ (0.1470), winning in 26 of 26 random seeds against Two-Stage (Table 1).",
        "file": "pages/claim-5/page.md",
        "children": []
      },
      {
        "slug": "conclusion",
        "title": "Conclusion",
        "file": "pages/conclusion/page.md",
        "children": []
      }
    ]
  }
}