Neural systems with range reducers and/or extenders

A neural system for processing an exogenous input process to produce an outward output process, said neural system comprising:

• a recurrent neural network;
• and
• at least one range extender,
• wherein at least one weight of said recurrent neural network is a nonlinear weight determined in a training of said neural system, and wherein said range extender is a dynamic transformer that transforms an output from at least one output neuron of said recurrent neural network and produces at least one component of said outward output process, said at least one component for a current time depending not only on a current value of said output from said at least one output neuron for said current time but also on at least one preceding value of said output from said at least one output neuron for a preceding second time prior to said current time.

An article of manufacture for training a neural system for processing an exogenous input process to produce an outward output process, said neural system comprising a recurrent neural network and at least one dynamic range transformer, wherein at least one weight of said recurrent neural network is a nonlinear weight determined by said training, said article of manufacture comprising:

• evaluation means for evaluating a training criterion;
• and
• adjusting means for adjusting the values of said neural system's weights to reduce the value of said training criterion,
• wherein an output of said range transformer depends not only on a current value of an input of said range transformer for a current time but also on a preceding value of said input of said range transformer for a preceding time prior to said current time.