Adaptive classifier having multiple subnetworks

DETAILED DESCRIPTION In accordance with the present invention, a neural network and methods of operation within such a network are disclosed.
The neural network is provided having an input layer, a middle layer and an output layer.
A method of operating the network includes the steps of presenting an input vector having a plurality of training features to the neural network, concurrently computing distances between a plurality of said training features and a plurality of prototype weight values and in response to an indication of a last training epoch, storing a count value corresponding to the number of occurrences of an input vector within a prototype.
With this particular arrangement a probabilistic neural network is provided.
The network may perform a plurality of pipeline operations to permit real time classification in response to applied input vectors.
The neural network may include a prototype memory having a plurality of weight values stored therein.
A distance calculation unit is coupled to the prototype memory and to a math unit.
The distance calculation unit may perform a plurality of concurrent calculations to compute a distance between an input vector and each of the plurality of weight values stored in the prototype memory.
The method and apparatus in accordance with the present invention may be fabricated as an integrated circuit device to permit classifications to be performed at high speed and thus permit utilization of the device for numerous classifications which were heretofore difficult to perform using conventional neural networks.
Furthermore, other learning algorithms may be seamlessly accommodated via a programmable resident microcontroller.
For example, a probabilistic neural network (PNN) may be incorporated into the network via an RCE procedure by setting the initial value of the middle-layer cell thresholds to zero.
This will cause the addition of a new middle-layer cell with the presentation of each new training pattern.
Thus the middle-layer cells act as storage elements for the full training set