1
Introduction
The concept proposed can be considered as the joint
between neural networks and artificial intelligence. Despite great
successes achieved in neural networks and neurocomputers during last
decade, the mapping of higher functions of brain onto the neural net
architecture remains the unsolved task. At the same time, in the area
of artificial intelligence the main problem is namely the absence of
intelligence in artificial systems (programs), i.e., the problem of
simulation of internal activity and inductive procedures. The proposed
concept pretends on the complex decision of these problems. Due to the
lack of available space this article practically does not contain
particular model equations and algorithms: these aspects of concept are
described in detail in [1].
The base of our concept is a dynamic threshold element,
many properties of which are borrowed from the real neuron. To
emphasize this, such an element hereinafter will refer to as neuron.
The important moment for further understanding of concept is that our
neuron is more complex than widely used models. The main output
parameter of our neuron is spike frequency, therefore the transfer
function of our neuron is gradual, but unlike popular sigmoid it has
not two, but three stable states. Further, our neuron has the property
of growing old, as far as we consider the neuron as a live unit.