Complexity of block-sequential update for symmetric neural networks
Eric Goles, Martı́n Matamala · 2005
We prove that the dynamics of arbitrary neural networks (not necessarily symmetric) of size n can be simulated by symmetric neural nets of size 3n updated in a block-sequential mode. As a particular case we prove that the class of symmetric neural nets with arbitrary diagonal elements updated sequentially is universal i.e. it simulates any nonsymmetric neural networks dynamics.