Final Dataset

The final dataset is different from the previous ones. Until now, the network learned to classify points as either BLUE or RED. Each point was described by two input values: its X and Y coordinates.

Recognizing handwritten digits is a common classification task with many practical applications, for example in document processing, postal services, and form recognition.

The new dataset contains images of handwritten digits from 0 to 9. Each image consists of 28 × 28 pixels. The goal is to train the network to recognize which digit is shown.

A 28 × 28 image contains 784 pixels, so the network now needs 784 input neurons, with one neuron representing each pixel.

Displaying all 784 input neurons would take up too much space. Instead, the input layer is represented by the digit image itself. Therefore, the first visible layer in the network is the first hidden layer.

The output layer now consists of 10 neurons. These are required since we no longer just decide between red and blue, but now distinguish between 10 different digits.