p. 18 (display before (24)) — the MSE of is
ProvedPrivAcq.Central.mse_linear_estimatorlaplace-mechanismlinear-estimatormean-square-errorp2o-batch-pfp2ap2o-gran-per-chapterp2o-plan-paperp2o-v1
On a probability space, let be identically distributed random variables with mean , variance and , and let be a random variable with law , , such that are mutually independent. Let , let , and let be weights with . Then the estimator has mean square error
This closed form is the MSE used throughout the mechanism design problem of the central setting; it is the formula mse of the model file.
Formalization Note Independence is stated for the family indexed by Option (Fin n) with none ↦ L and some i ↦ Z i; the law of is the published PrivLearn.Generic.laplace (1/η) (density ).
Preamble
import Mathlib import Definitions.Def_PrivLearn_Generic_Privacy
Formal statement
namespace PrivAcq.Central
open MeasureTheory ProbabilityTheory
theorem mse_linear_estimator {Ω : Type*} [MeasurableSpace Ω] (P : Measure Ω)
[IsProbabilityMeasure P] {n : ℕ} (θ var η : ℝ) (hη : 0 < η) (w : Fin n → ℝ)
(hsum : ∑ i, w i = 1) (Z : Fin n → Ω → ℝ) (L : Ω → ℝ)
(hZmeas : ∀ i, Measurable (Z i)) (hLmeas : Measurable L)
(hindep : iIndepFun (fun o : Option (Fin n) => Option.elim o L Z) P)
(hident : ∀ i j, IdentDistrib (Z i) (Z j) P P)
(hbdd : ∀ i ω, |Z i ω| ≤ 1 / 2)
(hmean : ∀ i, ∫ ω, Z i ω ∂P = 0)
(hvar : ∀ i, variance (Z i) P = var)
(hL : P.map L = PrivLearn.Generic.laplace (1 / η)) :
∫ ω, (∑ i, w i * (θ + Z i ω) + L ω - θ) ^ 2 ∂P = 2 / η ^ 2 + ∑ i, (w i) ^ 2 * var := by sorry
end PrivAcq.Central
Source
Fallah, Makhdoumi, Malekian, Ozdaglar, arXiv:2201.03968v2, p. 18 (display before (24)); restated in the proof of Theorem 3, p. 44; data model p. 7
Human review
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.