Extended Cauchy Machines

摘要

The Cauchy machine is a stochastic neural network related to the Boltzmann machine that uses fast simulated annealing instead of classical simulated annealing. Both these machines have the advantage that they enable a global minimum search of the error function. The main drawback of these machines is the slow training process and, like with other multilayer neural networks, the lack of knowledge for choosing the number of hidden neurons. If higher order connections are employed, a smaller number of training cycles can be expected and hidden neurons can be avoided. Hence, high order extensions of the Cauchy machine has been developed and are presented here.