EKI(G::Function, x_ensemble::Matrix{T}, y_target=zero(T), alpha=T(1e-8)*I)
solve w.r.t x model y = G(x) according to:
`G` : model that takes a Vector `x` of dimension `D` and gives a predicted observation Vector `y` of size `K`
`x_ensemble` : Matrix of size D-by-N, initial population of candidate solutions
`y_target` : Vector of size K, representing the target model observation
`alpha` : Covariance matrix of size K-by-K acting as a regularization
returns a new x_ensemble matrix with lower prediction error.