Circuit and method of error correcting with an artificial neural network

A method of performing bi-directional signal transformations through a network of weighting elements, comprising the steps of:

• processing elements of an input signal vector in a forward direction through a plurality of synapses respectively in each of a plurality of neurons for providing an output signal from each of said plurality of neurons;
• monitoring said output signals from said plurality of neurons to select at least one of said output signals from said plurality of neurons having a greatest activity level;
• generating a reverse flow signal corresponding to said selected at least one of said output signal having said greatest activity level;
• and
• processing said reverse flow signal in a reverse direction through said selected at least one of said plurality of neurons for updating said elements of said input signal vector.

A neural network, comprising:

• a plurality of neurons each having a plurality of first synapses coupled for receiving elements of an input signal vector as applied at a plurality of inputs, said plurality of neurons each having an output for providing an output signal in response to said input signal vector;
• a second synapse having an input coupled for receiving said output signal of a first one of said plurality of neurons and having an output coupled for summing with said output signal of said first one of said plurality of neurons, said second synapse having a predetermined positive value;
• and
• a third synapse having an input coupled for receiving said output signal of a second one of said plurality of neurons and having an output coupled for summing with said output signal of said first one of said plurality of neurons, said third synapse having a predetermined negative value.

A method of performing bi-directional signal transformations through a network of weighting elements, comprising the steps of:

• processing elements of an input signal vector in a forward direction through a plurality of synapses respectively in each of a plurality of neurons for providing an output signal from each of said plurality of neurons;
• feeding back a first one of said output signals from said plurality of neurons though a first synapse having a predetermined positive value to increase said first one of said output signals;
• and
• feeding back a second one of said output signals from said plurality of neurons though a second synapse having a predetermined negative value to decrease said first one of said output signals where a resulting said first one of said output signals is reverse processed through said plurality of synapses.