[Study notes]
Many people learned neural network for a long time, but there has been a fundamental problem plaguing himself. Why do the neural network to identify, judge, predict? Why the neural network can do it? In fact, we are in life, always unconsciously often using neural network. But we did not observe nothing. For example, often have friends ask us, where the house is roughly how much money? For example, the following example to determine prices, to know when to place in Beijing two years, donghuamen next to Tiananmen Square, the input x1 it should be equal to 50,000 yuan square meter. It is four floors, not high not low, an X2 equal to 800 dollars per square meter. If the floor is the top, it is too high, x2 is equal to 100 yuan a square meter. When the orientation is known after sunrise, X3 equal to 1000 dollars per square meter. If the orientation is known since the north, S3 equal to 100 yuan a square meter. But x1, x2, x3, these three parameters in your mind right weight is completely different, we all know, determine the price, location is the most important, therefore, the value of the weights w1 would be much higher, say 0.6. The floor of the heavy weights is not so high, for example, accounted for only 0.1. It accounted for 0.2 weight toward the south. Basic shift with the whole era have a great relationship, we are buy or not to buy, so when we want to buy a house, this offset will be high, such as b equal to one of thousands per square meter. Therefore, according to the following diagram: the final output: x1 * w1 + x2 * w2 + x3 * w3 + b = 50000 * 0.6 + 800 * 0.1 + 1000 * 0.2 + 10000 = 40280 yuan a square meter, the price.
Reprinted from the original article: https://blog.csdn.net/qq_43650923/article/details/100524839