Documentation
¶
Overview ¶
Program simple show cases a simple and contrived example of training a neural network to recognize data. I don't think its fair to say that this program shows learning or pattern recognition, because the network generated is extremely overly fitted to the data, since the data covers every possible input, but that is the nature of working with such a contrived example as this.
Now for the example details. This network takes in a 5x5 grid of input variables which contain 7 segment display style numbers encoded in them. The output of this network is intended to classify each input as a single digit number.
There are only 32 possible ways to represent these seven segment display encodings, and all 32 are used here, which is why this example is over fitted. This concept does scale up, however. If we were to make this a 200x200 grid of inputs, and loaded in examples of handwritten numbers, learning and pattern recognition could certainly arise from the proper training set.
Source Files