File size: 2,615 Bytes
943e43c | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 | {
"schema_version": 1,
"title": "Repro: Optimal Fair Aggregation of Crowdsourced Noisy Labels using Demographic Parity Constraints",
"emoji": "chart",
"space_id": "snaykey/repro-fair-crowd-aggregation",
"paper": {
"openreview_id": "niQUth28zn"
},
"tags": [
"icml2026-repro",
"paper-niQUth28zn"
],
"updated_at": "2026-07-27T00:00:00+00:00",
"root": {
"slug": "index",
"title": "Repro: Optimal Fair Aggregation of Crowdsourced Noisy Labels using Demographic Parity Constraints",
"file": "pages/index.md",
"children": [
{
"slug": "executive-summary",
"title": "Executive summary",
"file": "pages/executive-summary/page.md",
"children": []
},
{
"slug": "claim-1",
"title": "The demographic parity gap of majority-vote aggregated labels is bounded above by an exponentially decaying term in the number of annotators R, of the form sum over groups a of E[e^{-R K_phi(a,X)}] (Proposition 3.2).",
"file": "pages/claim-1/page.md",
"children": []
},
{
"slug": "claim-2",
"title": "Under interpretable conditions on annotator skill, both Majority Vote and Bayesian aggregation are shown to achieve asymptotically unbiased fairness gaps relative to ground-truth labels as crowd size grows (Theorem 3.4).",
"file": "pages/claim-2/page.md",
"children": []
},
{
"slug": "claim-3",
"title": "For majority voting with R annotators, the aggregated fairness gap is bounded by epsilon(R) times the sum of individual annotators' fairness gaps, showing majority voting can amplify rather than average out bias (Proposition 3.6).",
"file": "pages/claim-3/page.md",
"children": []
},
{
"slug": "claim-4",
"title": "Convergence of the fairness gap to the ground-truth value requires a majority of annotators to have skill exceeding 0.5 plus a margin epsilon; without this, majority voting fails to converge (Section 3, following Proposition 3.6).",
"file": "pages/claim-4/page.md",
"children": []
},
{
"slug": "claim-5",
"title": "FairCrowd, a post-processing method, enforces a strict epsilon-fairness constraint on the output of any aggregation rule while provably bounding the resulting accuracy loss (Theorem 4.1).",
"file": "pages/claim-5/page.md",
"children": []
},
{
"slug": "conclusion",
"title": "Conclusion",
"file": "pages/conclusion/page.md",
"children": []
}
]
}
}
|