Spiking neural network with pulse neural networks

(Original address: Wikipedia)

Introduction:

Pulse Neural Network spiking Neuralnetworks (Snns) is the third generation neural network model, the simulation neuron is closer to reality, besides, the influence of time information is considered. The idea is that neurons in a dynamic neural network are not activated in every iteration of the transmission (whereas in a typical multilayer perceptron network), but only when the membrane potential reaches a certain value. When a neuron is activated, it produces a signal that transmits to other neurons, raising or lowering its membrane potential.

In a pulsed neural network, the current level of activation of a neuron (modeled as a differential equation) is usually considered to be the current state, and an input pulse causes the current value to rise, for a period of time, and then gradually decay. There are many coding methods to interpret these output pulses as an actual number, which takes into account both the pulse frequency and the pulse interval time.

Preface:

Alan Lloyd Hodgkin and the Andrew Huxley in the 1952 The first Impulse neural network model is presented, which describes how the action potential is generated and transmitted. However, pulses are not transmitted directly between neurons, and it is necessary to exchange a chemical called "neurotransmitter" between synaptic gaps. The complexity and variability of this organism leads to the generation of many different neuron models.

from the point of view of information theory, it is a problem to find a model which can explain the impulse, that is, the action potential. Therefore, one of the fundamental problems of neuroscience is to determine whether neurons communicate through time coding. The time code indicates that a single neuron can replace hundreds of s -type hidden layer nodes.

Application:

The neural network can be used in information processing in general and the traditional artificial neural network, and the impulse neural network can model the problem of finding food for a virtual insect without prior knowledge of the environment. And, because it is closer to the reality of performance, so that it can be used to learn the work of the biological nervous system, electrophysiological pulses and pulse neural networks compare to the analogue output of a computer, which determines the likelihood of topological and bio-neurological hypotheses.

There is a major difference between the impulse neural network and the proven theory in practice. Pulsed neural networks have been shown to be useful in neuroscience systems, but not in engineering, and some large-scale neural networks have been designed to take advantage of pulse codes found in impulse neural networks, which are based on the principles of the reserve pool, but in reality, Large-scale impulse neural network computing because of the need for more computing resources and small capacity, development is limited, resulting in only a small number of large-scale impulse neural network is used to solve complex computational problems, which were previously resolved by the second generation of neural networks. It is difficult to add time to the second generation neural network model, and the Impulse neural network (especially when the algorithm is defined as discrete time) is quite easy to observe its kinetic characteristics. It is difficult to establish a model with stable behavior to achieve a specific function.

Spiking neural network with pulse neural networks