Hidden Layer

A layer that is not the output and not the input layer.

Estimate its size

With the Pearson Correlation we can see when two Neurons are doing the same job.

⟨fi,fj⟩=∑t(fi(t)−fˉi)(fj(t)−fˉj)∑t(fi(t)−fˉi)2∑t(fj(t)−fˉj)2\left\langle f_i, f_j\right\rangle=\frac{\sum_t\left(f_i(t)-\bar{f}_i\right)\left(f_j(t)-\bar{f}_j\right)}{\sqrt{\sum_t\left(f_i(t)-\bar{f}_i\right)^2 \sum_t\left(f_j(t)-\bar{f}_j\right)^2}}

When the highest Correlation is near one we have to reduce the size. Otherwise we can increase it.