Artificial neural network is like a black box, which is used to simulate any function. According to a certain training sample (that is, the input-output relationship of the required simulation function is known), the internal structure of the neural network can be changed to make its model characteristics approach the training sample. The so-called self-learning, self-organization and self-adaptation. Moreover, because the neural network adopts the method of global approximation, the overall model characteristics will not be affected by individual sample errors, which is called fault-tolerant characteristics.
In fact, it is easier to understand with bionic examples. Just like a baby, his parents keep teaching him to speak, and he can finally learn to understand the meaning of his parents' language. Occasionally, if his parents say a word or two wrong, the child can understand.