Say θ∈RD and ℓ(θ) is the log-likelihood. Then ∇2ℓ(θ) is the Hessian of the log-likelihood (the DxD matrix of the partiel second derivatives for θ). The observed Fisher information is I* = −∇2ℓ(θ∗) with θ∗ being the vector of the maximum likelihood estimates. The standard errors are the square-roots of the diagonal of the inverse Fisher Information:
se = sqrt( diag(I∗−1) )
Example: exponential distribution with parameter θ
L(x) = θ·exp(-θ·x)
ℓ(x) = log(θ) - θ·x
ℓ(x) = n·log(θ) - θ·Σixi (for a sample of n values x = (x1, x2, ..., xn))