Associative-Projective Neural Networks (APNNs) - Neural Networks and Micromechanics - page 75

Fig. 5.2 Functional diagram
of the neuron

the additional output of the neuron. The set inputs S 1 , ... , S k are joined to the OR

element. The inhibitory inputs R 1 , ... , R k are joined to another OR element. The

additional output of neuron Q * is used to control the activation of the associative

field described below.

5.1.2 Neural Fields

The basic structural unit of the proposed associative-projective neural network is a
neural field [ 2 , 5 ]. The neural field is the subset of the neurons, which carry out
identical functions. For the neurons of one neural field, all the connections with the
neurons of this field and the neurons of other fields are formed according to the rules
common for all neurons of the field. We will consider that a quantity of neurons in
each neural field equals the same value n . We distinguish between two types of
neural fields: associative and buffer. The associative neural field is designated for
the formation of neural ensembles during the neural network training and for the
associative restoration of neural ensembles in the mode of recognition. The buffer
neural field is designated for the temporary storage of ensembles, for the normali-
zation of the sizes of neural ensembles, and for the auxilliary operations with the
ensembles. With the aid of the buffer neural field, it is possible to determine the
difference between two excited neural ensembles. We also will distinguish between
different functions that can realize the buffer field. Any neural field can have the
special device, which we will call the counter, that determines a quantity of excited
neurons in this field.
Associative neural field .
Let us examine in detail the functions and properties of the associative neural
field. The associative field is represented in Fig. 5.3 . Like any other field, it contains
n neurons, whose outputs are the outputs of the neural field. Furthermore, each of
the outputs is supplied to all associative inputs of the neurons of the same field.
Thus, the neurons are connected to each other with associative connections. The
reinforcement signal tr is supplied to the training inputs of all the neurons of the