§5, p. 20 — β*_N(γ) = β_N(F − f*, γ)
ProvedLearnNoConc.ERM.betaStar_eq_betaNp2o-batch-pfp2ap2o-gran-per-chapterp2o-plan-paperp2o-v1rademacher-complexitystatistical-learning
Let be a probability space, a class of real functions on , , and . Then the parameter of Definition 2.1 equals the parameter of Definition 5.2 for the shifted class :
The two definitions differ only in normalization ( with the bound , versus with the bound ) and in the centre of the ball. The identity transfers the results proved for general classes in §5 to the parameter that appears in Theorem 3.1.
Formalization Note Both sides are -valued infima, with expectations of suprema taken in .
Preamble
import Mathlib import Definitions.Def_LearnNoConc_ERM_Setting
Formal statement
namespace LearnNoConc.ERM
open MeasureTheory
open scoped ENNReal
/-- §5, p. 20: `β*_N(γ) = β_N(F − f*, γ)`. -/
theorem betaStar_eq_betaN {Ω : Type*} [MeasurableSpace Ω] (μ : Measure Ω)
[IsProbabilityMeasure μ] (F : Set (Ω → ℝ)) (fstar : Ω → ℝ) (N : ℕ) (hN : 0 < N)
(γ : ℝ) (hγ : 0 < γ) :
betaStar μ F fstar N γ = betaN μ (shift F fstar) N γ := by sorry
end LearnNoConc.ERM
Source
Mendelson, Learning without Concentration, arXiv:1401.0304v2, §5, p. 20, "observe that β*_N(γ) = β_N(F − f*, γ)"
Human review
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.