An Empirical Investigation of Model-to-Model Distribution Shifts in Trained Convolutional Filters