Neurons Tune Their Own Excitability When They Make A Decision

The brain is usually considered to be an extraordinarily complex object while a neuron is believed an ideal logical element. This idealization does not correspond to the real neural system properties. Experimental data give trustworthy evidence that a neuron is a microcosm of the brain. The firing threshold seems to be a dynamic property of an excitable membrane. Neurons appear to evaluate the most preferable consequence of their participation in the brain action, transiently change their excitability and only after that compare the magnitude of the input signal and threshold. This suggests a novel approach to understanding the neuron’s role in controlling behavior. The model of a neuron has been developed on the basis of the supposed intracellular chemical reactions that are specific for a given input signal. The yield of chemical reactions was considered to affect the transition of sodium channels into an open state, as described in the Hodgkin-Huxley model, and to modulate neuronal excitability. However, for the given combination of inputs, the neuron model exhibits an all-or-none principle of spike generation when efficacy of the inputs changes. An increase in complexity of the neuronal model is more than compensated for by simplification of neural network tuning. 1 Necessity to modify the model of a neuron Conventional neural network models based on a simple neuron related to the McCalloch and Pitts model [1] have many well-known remarkable properties and can solve some problems in line with algorithms of the networks operation. This approach is easier to analyze than the behavior of complex neuron units. However, irrespective of the physiological relevance of these models, they have some disadvantages [2]. The form of algorithms in general is determined by the model of the neuron, by the network structure and by the salient feature of the practical problem. The tuning is usually slow and sometimes accompanied by a network paralysis. Memory capacity is low and huge network creations are necessary. False images may appear and it is difficult to overcome local minima. Properties of neural networks depend both on the network structure and on a model of neurons that is incorporated within the network. An artificial neuron is usually considered to make a summation of excitations and to generate an output reaction in accordance with its activation function. Neural networks consisting of such neurons store information, thus changing the efficacy of synaptic connections between neurons (Fig. 1). CONDITIONED UNCONDITIONED REACTION UNCONDITIONED CONDITIONED REACTION NETWORK NEURON