Definition 3 — the feasible region
Definitionsu_feasibleRegionmeasure-theoryprobabilitystatistics
Let be a family of candidate distributions indexed by a parameter , let be an unwanted and a retained member of the family, and let be baseline trade-off functions. The feasible region is the set of parameters whose distribution satisfies -distributional unlearning with respect to and :
It is the set of edited distributions that simultaneously erase the unwanted domain to the prescribed level and preserve the retained one. In Lean this is StatUnlearn.feasibleRegion fam p₁ q₁ f_d f_c, a Set of parameters.
Definition code
/-
Mission: Statistical Unlearning of Distributions
Source: A. Pandey, S. Kulkarni, arXiv:2605.16645 — Definition 3 (the feasible region).
-/
import Mathlib.MeasureTheory.Measure.MeasureSpaceDef
import Mathlib.Data.Real.Basic
import Definitions.Def_su_unlearning
open MeasureTheory
namespace StatUnlearn
/-- **Feasible region** (Definition 3 of the paper).
For a family of probability distributions `fam` indexed by `ι` (the paper's class
`𝒫`), unwanted/retained members `p₁, q₁ : ι`, and baseline trade-off functions
`f_d, f_c`, the feasible region is the set of indices `μ` whose distribution satisfies
`(f_d, f_c)`-unlearning with respect to `fam p₁` and `fam q₁`. -/
def feasibleRegion {W : Type*} [MeasurableSpace W] {ι : Type*} (fam : ι → Measure W)
(p₁ q₁ : ι) (f_d f_c : ℝ → ℝ) : Set ι :=
{μ | IsUnlearning (fam μ) (fam p₁) (fam q₁) f_d f_c}
end StatUnlearn
Source
Pandey, Kulkarni, "Statistical Unlearning of Distributions: A Hypothesis Testing Approach", arXiv:2605.16645, p. 4, Definition 3