Simultaneous estimator confidence for the concrete adaptive policy
ProvedLG2021.Concentration.estimatorGood_failure_leLet be positive integers, let , and let . Run the concrete policy with the published configuration and candidates on a measurable legal set with a legal fallback anchor. Assume the unknown loss has values in on and the actual observation kernel satisfies the normalized conditional noise condition after every legal history. For any comparison points, the published simultaneous scalar and directional-gradient confidence event satisfies
under the actual history law of the first observations. This combines the two stage-confidence failure bounds with the explicit logarithmic confidence choice. Its event and probability law are those of the actual policy and seed distribution.
This is a formalization auxiliary for Algorithms 1–2 and their probability analysis in Tor Lattimore and András György, Improved Regret for Zeroth-Order Stochastic Convex Bandits, COLT 2021, PMLR 134, Lemma 4 and Sections 5–7. The implementation and event predicates are published definitions; the guarantee is proved in this theorem. Adaptive observation kernels and exact finite-history semantics are explicit formal interfaces. Statistical concentration and regret claims are not assumed by the implementation definitions.
import Mathlib import Definitions.Def_LG2021_Policy open LG2021 LG2021.Algorithm LG2021.Concentration BilateralProfitMHR MeasureTheory ProbabilityTheory Set Metric
theorem LG2021.Concentration.estimatorGood_failure_le {d n u N : ℕ} (hd : 0 < d) (hn : 0 < n) (hu : 0 < u)
{D : ℝ} (hD : 0 ≤ D) (K : Set (Space d)) (hK : MeasurableSet K)
(R : ℝ) (anchor : Space d) (ha : anchor ∈ K)
(ν : ObsKernel (Space d)) (hνm : ν.IsMeasurable) (F : Space d → ℝ)
(hν : CorollaryTransport.UnitNoiseCond K F ν) (hF : ∀ x ∈ K, F x ∈ Set.Icc (0 : ℝ) 1)
(hN : N ≤ n) (v : Fin (d+1) → Space d) :
(historyMeasure (policy (u := u) (oracleConfig d D n) K hK R anchor)
(seedMeasure d u (oracleConfig d D n)) ν hνm (2*N)).real
(estimatorGood (oracleConfig d D n) K hK R anchor ν hνm N v)ᶜ ≤ 3 * (d+2) / (n : ℝ) := by sorry