The basic characteristics of artificial neural network

The ① Artificial Neural Network (ANN) is a widely connected giant system. Neuro-scientific research shows that the main part of the human central nerve cortex is composed of 10[11]~10[12] neurons, each neuron has a 10[1]~10[5] synapse, Synapse is a junction between neurons, determining the strength and nature of the connection between neurons. This suggests that the cerebral cortex is an extensively connected giant complex system, and Ann's connection mechanism mimics this characteristic of the human brain.

② Artificial Neural Network (ANN) has its parallel structure and parallel processing mechanism. Ann is not only structurally parallel, but also parallel and simultaneous in its processing order. The processing unit in the same layer is operated simultaneously, that is, the computational function of neural network is distributed on multiple processing units.

The distributed structure of ③ artificial neural Network (ANN) makes it have the same fault tolerance and associative ability as the human brain. The brain has a strong fault tolerance. We know that every day brain cells die, but not affect people's memory and thinking ability. This is precisely because the brain's storage of information is realized by changing the synaptic function, the information is stored in the distribution of the neuronal connection intensity, the storage area and the operation area are merged, the different information is naturally communicated, and its processing is a large-scale continuous-time mode. and the acquisition of storage knowledge using the "association" approach. This is similar to human and animal associative memory, when a neural network input an excitation, it is in the stored knowledge to find and input matching the best storage knowledge for its solution.

④ Artificial Neural Network (ANN) has self-learning, self-organization and adaptive ability. Brain function is restricted by congenital factors, but the acquired factors (such as experience, learning and training) also play an important role. Ann simulates this characteristic of the human brain well. If the final output is incorrect, the system can adjust the weights added to each input to produce a new result, which can be achieved by a certain training algorithm. The training process is complex, with data being repeatedly entered through the network, and the weights are adjusted each time to improve the results, eventually achieving the desired output. The network has gained experience in the course of training. Theoretical studies show that Ann is able to achieve any continuous mapping by choosing the right Ann, and the ability of classification, generalization and association is demonstrated by the learning of samples.

The basic characteristics of artificial neural network