Brain.js 示例

Question

I decided to get known with a neural network in js. I took brain.js library as an example. This is taken from their GitHub page. I can understand what that we have two input neurons 1 and 0. do I get also two output 1and 0? I read about the weights, where are they here and i don't really understand what are they? In output what do we pass?

var net = new brain.NeuralNetwork();

net.train([{input: [0, 0], output: [0]},
           {input: [0, 1], output: [1]},
           {input: [1, 0], output: [1]},
           {input: [1, 1], output: [0]}]);

var output = net.run([1, 0]);  // [0.987

Answer 1

This is a simple neural network that approximates the XOR logic operator. The result of the network will be in range from 0 to 1.

The result will (mathematically) approach 1 the larger the training set gets. Because this is a simple example the value predicted is close enough.

Usually, one will use a threshold/stepping function like the sigmoid function, to transform the output from linearly separable to approximately binary.

Here's a possible, simple NN for XOR approximation: