An input vector h = [1, 5, 7] is passed through a normalization layer. The layer computes the output using the formula α * (h / (sqrt(mean(h^2)) + ε)) + β. Given a learnable gain parameter α = 1.5, a learnable bias parameter β = 0.5, and a numerically stabilizing constant ε that is small enough to be ignored in this calculation, what is the resulting output vector?

A machine learning engineer is comparing two normalization functions for a neural network layer. The input is a vector h, and ε is a small constant for numerical stability.

Function A: output = gain * ((h - mean(h)) / (std_dev(h) + ε)) + bias Function B: output = gain * (h / (root_mean_square(h) + ε)) + bias

What is the primary consequence of Function B omitting the subtraction of the input's mean (- mean(h)), a step which is present in Function A?