This reminds me of dropout regularization where you improve a neural network by removing random neurons from the neural network.

See https://en.wikipedia.org/wiki/Dropout_(neural_networks) and https://youtu.be/u4alGiomYP4?t=33m53s