Neural network: Realization of Perceptron and linear neural network

Tips: This article is a reference to the mechanical industry press "neural network Design" (Dai Qu, etc.) a book compiled by the relevant procedures, for beginners or want to learn more about the neural network kernel enthusiasts, this is the most reading value of the textbook.

Perceptual machines and linear neural networks are the simplest and most basic types of neural networks, but they are also widely used. The learning rules of perceptron and linear neural networks are quite simple, for the Perceptron:

For linear neural networks there are:

It can be seen that there are many similarities between the two, so the common functions of the two can be put in a module in programming, but the transfer function of the perceptron is the hard limit transfer function, and the linear network is the linear function.

In the upper, W is the weight matrix, p is our input vector, E represents the expected value and the actual output error, B is the bias.

It has been proved theoretically that, as long as the solution of the weights is present, the Perceptron and the linear network always converge, but in what case the solution of the weight is present, which is the limitation of the Perceptron: he can only divide the linear and measurable objects. Of course we're talking about a single layer of perceptual machines.

In many of today's software, there are a lot of tools to provide a neural network toolbox, the most convenient for a certain too matlab, these tools easy to use, fast, but they are the implementation of the process of encapsulation at the bottom, and because of the large size of the toolbox, in some practical projects we can not directly transplant him, So the realization of its own process can not only experience the charm of knowledge, but also to the book knowledge can be more profound understanding, why not?