(Bennett's inequality) This exercise is devoted to a proof of a strengthening of Bernstein's inequality, known...

 

Transcribed Image Text:

(Bennett's inequality) This exercise is devoted to a proof of a strengthening of Bernstein's inequality, known as Bennett's inequality. (a) Consider a zero-mean random variable such that IX; ≤ b for some b > 0. Prove that 108 E[et] ≤ 0²1² (0 (1 = 26} -1-Ab = for all λ € R, (ab)² where o var(X;). = (b) Given independent random variables X₁,..., X, satisfying the condition of part (a), let ²:= be the average variance. Prove Bennett's inequality n P[ΣX, 2 no] ≤ exp{-10² h (55)}. s X₁ i=1 where h(t) = (1 +t) log(1 +t) - t for t > 0. (c) Show that Bennett's inequality is at least as good as Bernstein's inequality. (2.62) (Bennett's inequality) This exercise is devoted to a proof of a strengthening of Bernstein's inequality, known as Bennett's inequality. (a) Consider a zero-mean random variable such that IX; ≤ b for some b > 0. Prove that 108 E[et] ≤ 0²1² (0 (1 = 26} -1-Ab = for all λ € R, (ab)² where o var(X;). = (b) Given independent random variables X₁,..., X, satisfying the condition of part (a), let ²:= be the average variance. Prove Bennett's inequality n P[ΣX, 2 no] ≤ exp{-10² h (55)}. s X₁ i=1 where h(t) = (1 +t) log(1 +t) - t for t > 0. (c) Show that Bennett's inequality is at least as good as Bernstein's inequality. (2.62)