Prove2Me
Navigate
DiscoverCollectionsFormalpediaGalaxyBlogsUsersMomentumMy Missions+
Prove2Me
Log in
← Formalpedia

Exp4 estimated-advantage bound

Proved
BanditAlgorithm.exp4_estimate_advantage_bound

by Harry_Xu · Jul 28, 2026 · Mathlib 0df444a (Lean v4.33.1)

bandit-algorithmsonline-learning

Let k≥1k\ge 1k≥1, M≥2M\ge 2M≥2, and n≥1n\ge 1n≥1, and let the learning rate satisfy η>0\eta>0η>0. In round ttt, let xt,a∈[0,1]x_{t,a}\in[0,1]xt,a​∈[0,1] be the reward of arm aaa, and let each expert advice row (Et,m,a)a(E_{t,m,a})_a(Et,m,a​)a​ be a probability distribution over the arms. Suppose the learner follows Exp4 with learning rate η\etaη and exploration parameter γ=0\gamma=0γ=0.

For every fixed expert mmm, write S~n,m\widetilde S_{n,m}Sn,m​ for its cumulative importance-weighted reward estimate and ∑t=1nXt\sum_{t=1}^{n}X_t∑t=1n​Xt​ for the learner’s collected reward. Then

Ex,π ⁣[S~n,m]−Ex,π ⁣[∑t=1nXt]≤log⁡Mη+ηnk2.\mathbb{E}_{x,\pi}\!\left[\widetilde S_{n,m}\right] - \mathbb{E}_{x,\pi}\!\left[\sum_{t=1}^{n}X_t\right] \le \frac{\log M}{\eta}+\frac{\eta nk}{2}.Ex,π​[Sn,m​]−Ex,π​[t=1∑n​Xt​]≤ηlogM​+2ηnk​.

This is the reusable potential-and-second-moment estimate for Exp4 before unbiasedness is substituted and the maximum over comparator experts is taken.

Formalization Note The two expectations are integrals over adversarialMeasure; exp4Estimate denotes S~n,m\widetilde S_{n,m}Sn,m​.

Preamble
import Definitions.Def_ContextualAdversarialBandit
import Definitions.Def_exp4Policy

open MeasureTheory ProbabilityTheory
Formal statement
theorem BanditAlgorithm.exp4_estimate_advantage_bound
    {k M : ℕ} (hk : 0 < k) (hM : 1 < M) (n : ℕ) (hn : 0 < n)
    (η : ℝ) (hη : 0 < η)
    (x : ℕ → Fin k → ℝ) (hx : ∀ t : ℕ, ∀ a : Fin k, x t a ∈ Set.Icc (0 : ℝ) 1)
    (E : ℕ → Fin M → Fin k → ℝ)
    (hE0 : ∀ t : ℕ, ∀ m : Fin M, ∀ a : Fin k, 0 ≤ E t m a)
    (hE1 : ∀ t : ℕ, ∀ m : Fin M, ∑ a, E t m a = 1)
    (π : BanditPolicy k)
    (hπ : IsExp4Policy η 0 E π)
    (m : Fin M) :
    (∫ h, exp4Estimate η 0 E n h m ∂(adversarialMeasure x π n)) -
        (∫ h, (∑ t, (h t).2) ∂(adversarialMeasure x π n)) ≤
      Real.log M / η + η * ((n : ℝ) * k) / 2 := by
  sorry
Source
Lattimore and Szepesvári, Bandit Algorithms (Cambridge University Press, 2020), Theorem 18.1 proof, printed p. 230, Lemma 18.2 and Eqs. (18.9), (18.11), and (18.12). https://tor-lattimore.com/downloads/book/book.pdf

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