Towards a More Complete Theory of Function Preserving Transforms

Michael Painter

In this paper, we develop novel techniques that can be used to alter the architecture of a neural network, while maintaining the function it represents. Such operations are known a…