In this thesis, the reduction of neural networks is studied. A new, largely automatic method is developed for reducing a neural network, and its capabilities are analyzed and compared with current methods. An example is presented that is irreducible via existing techniques, but is reducible by the new method. Factors...