input value - a numeric value that is submitted to a neural network for processing.
input vector - a set of input values that form a unit record of processing data. These records are used for training the neural network.
output value - a numeric value that is produced by a neural network after processing the input vector. There can be more than one.
output vector - a set of output values that form a unit record of output data. These are the numeric results of a neural network that are used by the application that required the neural network.
supervised learning - learning method that requires a human to supply the desired output value(s) for the neural network to use during training.
unsupervised learning - learning method that does not require a human to supply the desired output value or values during training of the neural network. The neural network itself, during training, figures out what the output value(s) should be. Two good examples of network topologies that use unsupervised learning network algorithms are Kohonen networks and Learning Vector Quantization networks. These networks automatically cluster input data through competing hidden elements whose vector element values, over time, position the element at the center of the cluster they grew to represent.
neurode - a single processing element. A neurode usually has several input connections and one output connection.
input layer - the processing layer of a neural network that receives the data values that are part of an input vector.
hidden layer - any processing layer of a neural network that has no direct connection to outside the neural network. Any connection into this layer is from another hidden layer or the input layer. Any connection out from this layer is to another hidden layer or to the output layer.
output layer - the processing layer of a neural network that produces the output vector of values, that results after the neural network has finished processing a input vector of values.
training - the phase of neural network development where the training data, in the form of input vectors, is presented repetitively to the neural network so it can learn from that data.
|
