Applying the finite element method to neural networks for supercomputer scale

By applying the finite element method to neural networks supercomputer scale machine learning can done.

In fact, supercomputer scale FULLY AUTOMATED machine learning.

The finite elements would be autoencoders and logistic regression. Bulk synchronous parallel graph database such as Pregel or Graphlab would be used for greatly simplify coding. Graph partitioning software Chaco would be used to partition the graph (as is often done in finite element on supercomputers).

It would allow for a larger scale version of this work done by Jeff Dean of Google. http://techtalks.tv/talks/57639/