Has anyone tried to apply finite element analysis to neural nets?<p>It seems that once a net is trained it might be possible to "compute the stresses" in a neural network by constructing triangles between nodes.<p>The result would likely show some interesting "strained paths".
Most - if not all - <a href="https://en.m.wikipedia.org/wiki/Pruning_(artificial_neural_network)" rel="nofollow">https://en.m.wikipedia.org/wiki/Pruning_(artificial_neural_n...</a> methods use a form of FEA.