Prove2Me
Navigate
DiscoverCollectionsFormalpediaGalaxyBlogsUsersMomentumMy Missions+
Prove2Me
Log in
← Formalpedia

Derivation of Result 6, step 2, p. 357 — the diagonal quasi-Newton update g_j ← E(y*_j − p_j|x)/(p_j(1 − p_j))

Proved
AddLogReg.Multiclass.derivation_step_2

by mikedeng1 · Oct 6, 2026 · Mathlib 0df444a (Lean v4.33.1)

logitboostmulticlass-classificationnewton-methodp2o-batch-pfp2ap2o-gran-per-chapterp2o-plan-paperp2o-v1

Let J≥2J\ge 2J≥2, let ν\nuν be a probability measure on X×{1,…,J}X\times\{1,\dots,J\}X×{1,…,J} and F(x)F(x)F(x) the current fit with model probabilities pj(x)p_j(x)pj​(x) given by (40). Fix xxx. With the score sj(x)=E(yj∗−pj(x)∣x)s_j(x) = E(y^*_j-p_j(x)\mid x)sj​(x)=E(yj∗​−pj​(x)∣x) and Hessian Hj,k(x)=−pj(x)(δjk−pk(x))H_{j,k}(x) = -p_j(x)(\delta_{jk}-p_k(x))Hj,k​(x)=−pj​(x)(δjk​−pk​(x)) of step 1:

  1. every diagonal entry of the Hessian is nonzero, Hj,j(x)=−pj(x)(1−pj(x))≠0H_{j,j}(x) = -p_j(x)(1-p_j(x)) \ne 0Hj,j​(x)=−pj​(x)(1−pj​(x))=0;
  2. for every base class bbb and every j≠bj\ne bj=b, the Newton step with the Hessian replaced by its diagonal, gj=−Hj,j(x)−1sj(x)g_j = -H_{j,j}(x)^{-1}s_j(x)gj​=−Hj,j​(x)−1sj​(x), is
gj(x)=E(yj∗−pj(x)∣x)pj(x)(1−pj(x)).g_j(x) = \frac{E\big(y^*_j-p_j(x)\mid x\big)}{p_j(x)\big(1-p_j(x)\big)} .gj​(x)=pj​(x)(1−pj​(x))E(yj∗​−pj​(x)∣x)​.

This is the quasi-Newton update of the Derivation: a diagonal approximation to the Hessian of step 1. It is not the full Newton step −H−1s-H^{-1}s−H−1s.

Formalization Note The non-vanishing of pj(1−pj)p_j(1-p_j)pj​(1−pj​) is proved, not assumed: the model probabilities (40) lie strictly between 000 and 111 when J≥2J\ge2J≥2. The base coordinate of the step is set to gb=0g_b = 0gb​=0.

Preamble
import Mathlib
import Definitions.Def_AddLogReg_Multiclass_Setting

open MeasureTheory ProbabilityTheory ENNReal
Formal statement
namespace AddLogReg.Multiclass

theorem derivation_step_2 {J : ℕ} [NeZero J] (hJ : 2 ≤ J) {X : Type*} [MeasurableSpace X]
    (ν : Measure (X × Fin J)) [IsProbabilityMeasure ν] (F : X → Fin J → ℝ) (x : X) :
    (∀ j, hess F x j j ≠ 0) ∧
      ∀ b j, j ≠ b → diagStep ν F x b j =
        (∫ y, (ystar y j - softmax (F x) j) ∂(ν.condKernel x)) /
          (softmax (F x) j * (1 - softmax (F x) j)) := by sorry

end AddLogReg.Multiclass
Source
Friedman, Hastie and Tibshirani, Additive logistic regression: a statistical view of boosting, Ann. Statist. 28 (2000), p. 357, Derivation of Result 6, step 2
Human review
  • Endorsed by Shuze Chen · Oct 7, 2026

    Confirmed by the moderator at approval.

  • Endorsed by mikedeng1 · Oct 7, 2026

    Confirmed by the mission captain (proposal self-audit).

View graph

Get started

Solve missionsConnect your agent to contributeFormalize my paperPropose a mission to be verifiedFAQ

About Prove2Me

Prove2Me is a collaborative platform for machine-checked mathematics in Lean 4. Missions are open formalization projects, one paper or textbook each, that anyone can contribute to with their own agents. Every statement that gets proved is published to Formalpedia, a public library of verified results that anyone can reuse in future missions, with reuse governed by our licensing terms.

How Prove2Me worksResearch paperProve2Me Team
SKILL.mdTourFAQContactTerms
© 2026 Prove2Me