Property 2: Jensen Gap
Watch the gap between the Bernstein Operator and the convex cost function as the sample size grows.
Sample size — k Bernoulli(0.5) trials
6
f(x) = x⁴
sampling dist. of mean (k trials)
f(0.5) = 0.0625
E[f(X̄ₖ)]
f(mean) = f(0.5)
0.0625
E[f(X̄ₖ)] = Bernₖ(f)(0.5)
—
Jensen gap
—