Why hasn't the finite element method been applied to neural networks?

metaoptimize.com