# Эксперимент: Conditional OT barycenter feature augmentation (#463)

{
  "worked": true,
  "confidence": 7,
  "verdict": "Built a readable latent conditional-OT barycenter augmenter using Sinkhorn transport, gated source masses, transported soft labels, and a detached barycenter loss. The three toy predictions were confirmed: convex-hull violation was 0, consensus variance matched sigma^2/K within 0.99%, and gate-weight movement had slope -2.0 as predicted. However, augmentation did not improve held-out accuracy in this toy task: ERM and Mixup reached 0.924 versus 0.918 for OT, so there is no evidence of a practical accuracy win here.",
  "metrics": {
    "baseline": "ERM: target accuracy 0.9240, source accuracy 0.9479",
    "idea": "Conditional latent OT: target accuracy 0.9180, source accuracy 0.9375; mechanism checks passed with maximum variance relative error 0.00986, convex-hull violation 0, and gate slope -2.0"
  },
  "how_to_run": "/home/maxwelhelp/main/bin/python3 ot_barycenter_experiment.py",
  "files": [
    "ot_barycenter_experiment.py",
    "results.json"
  ],
  "limitations": "Only a small synthetic domain-shift classification task was tested, not PACS or Office-Home. The variance and gate-linearity checks use idealized consensus constructions rather than estimating those properties from the full entropic OT solver; end-to-end gradients, domain separability, rare-context accuracy, runtime overhead, CUDA execution, and multi-seed significance were not tested."
}