The compensated normalized variable satisfies the scalar interface on the alive window
ProvedG26G.Sharp.runN_SISetting. is a compact action set spanning , is a compact generator (), is the direction enumeration, and is a GIA-FTRL run of rounds (IsGIAFTRLRun) with cumulative loss estimate , run filtration , learning rate (gbeta) and magnitude constant (gB). The filtration used below is for and (Fx). The standing hypotheses Sharp hold: lies in the unit ball, , , is the unique minimizer of on , and the instance is sharp with modulus (H.sig): on . The corruption is bounded by : a.s. (IsCorruptedG). The projective body is and the cone body is ; , , and is the index of among the reference atoms (js).
Assume is surjective. Fix , a buffer with , , , , and assume for . Consider
and the alive events (IN): the reference was strictly at every round of , and for all , as in (G.523). Then holds for with thresholds , budgets and .
These are the variables whose local moments (G.568) are bounded on the alive window. Lemma G.26 then contracts them uniformly over .
Formalization Note Rounds are 0-based.
import Mathlib
import Definitions.Def_IAFTRL_LinEnv
import Definitions.Def_IAFTRL_GIAFTRL
import Definitions.Def_IAFTRL_G26_body
import Definitions.Def_IAFTRL_G26_run
import Definitions.Def_IAFTRL_G26_window
open MeasureTheory ProbabilityTheory Set IAFTRL Matrix
open G26G
open G26 G26R
variable {d : ℕ} {𝒜 ℬ : Set (Fin d → ℝ)} {e : ℕ → Fin d → ℚ} {θ astar : Fin d → ℝ}
variable {Ω : Type*} [mΩ : MeasurableSpace Ω] {P : Measure Ω} [IsProbabilityMeasure P]
{T : ℕ} {R : GRun d Ω}
open G26G.Sharp
variable (H : Sharp 𝒜 ℬ θ astar)
include Htheorem G26G.Sharp.runN_SI (he : Function.Surjective e) (hR : IsGIAFTRLRun P 𝒜 ℬ e T R) {C : ℝ}
(hcor : IsCorruptedG P 𝒜 θ C T R) {u : Fin d → ℝ} (hu : u ∈ DT ℬ θ astar) {τ : ℕ → ℝ}
(hτpos : ∀ n, 0 < τ n) (hτmono : ∀ n, τ n ≤ τ (n + 1)) (hδ : ∀ n, τ (n + 1) - τ n ≤ 1)
{m N : ℕ} (hN : N ≤ T)
(hτB : ∀ n, m ≤ n → gB d * (gbeta d (n + 1) * H.sig⁻¹ + 2 * H.sig⁻¹ + 2) ≤ τ n) :
SI P (Fx R T) (YN R θ H.sig⁻¹ C τ u) (IN 𝒜 ℬ e (js ℬ e astar) R θ astar H.sig⁻¹ C τ m) m N τ
(fun n => H.kb * gbeta d (n + 1)) (gB d) := by sorry