The Fisher information in a univariate model is given by:
I(θ)=−EY|θ[∂2∂θ2logp(y|θ)]
for data Y. In a multivariate model, the Fisher information matrix, has ij entry:
Iij(θ)=−EY|θ[∂2∂θi∂θjlogp(y|θ)]