Proof of Theorem 2, pp. 30–31 — the inner supremum for the hinge loss
ProvedRWPI.Classif.hinge_inner_supLet satisfy . Let , let and let . Then
This is the dual integrand of Proposition 1 for the hinge loss under the label-preserving cost , written in the displacement . It is the step of the proof of Theorem 2 that produces the penalty for the support vector machine.
Formalization Note The paper's display writes the transport norm as and the threshold as (the reverse of Theorem 2's statement; both readings are true since are conjugate); the statement uses Theorem 2's convention. Typographical artefacts of the page (the index for , a stray in the first case) are not reproduced. The supremum is taken in with truncated subtraction, harmless because the term at dominates every truncated term.
import Mathlib import Definitions.Def_RWPI_Classif_losses open scoped ENNReal
namespace RWPI.Classif
/-- App. A.1, proof of Theorem 2, pp. 30–31: for a label `y₀ ∈ {−1, +1}`, a point `x₀ ∈ ℝ^d`
and `λ ≥ 0`, `sup_Δ { (1 − y₀ βᵀ(x₀ + Δ))⁺ − λ ‖Δ‖_q } = (1 − y₀ βᵀx₀)⁺` if `‖β‖_p ≤ λ` and
`+∞` otherwise, where `1/p + 1/q = 1`. Stated in Theorem 2's convention (transport norm `ℓ_q`,
regularizer `ℓ_p`); the proof's display has the two exponents swapped. -/
theorem hinge_inner_sup {d : ℕ} (p q : ℝ≥0∞) (hpq : p.HolderConjugate q)
(β x₀ : Fin d → ℝ) (y₀ : ℝ) (hy₀ : y₀ = 1 ∨ y₀ = -1) (lam : ℝ) (hlam : 0 ≤ lam) :
(⨆ Δ : Fin d → ℝ, (ENNReal.ofReal (hingeLoss β (x₀ + Δ, y₀)) -
ENNReal.ofReal lam * ENNReal.ofReal ‖WithLp.toLp q Δ‖)) =
if ‖WithLp.toLp p β‖ ≤ lam then ENNReal.ofReal (hingeLoss β (x₀, y₀)) else ⊤ := by sorry
end RWPI.Classif
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.