Task 2

In this task, you will investigate a Feedforward Neural Network for classifying simple characters and then modify it manually.

Exercise 1 – Classifying Characters

Open the character classification tool.

Open the character classification tool

Step 1

Load network

Download the file 3x3network.json and open it in the tool using Load Net.

Step 2

Test characters

Draw the symbols

one after another, convert them to the grid representation, and test whether the network classifies them correctly.

Also investigate slightly modified drawings of the four symbols. Briefly describe when the network no longer recognizes a character correctly.

Exercise 2 – Modifying the FNN

Develop a network that can classify 5 different shapes using three helper shapes in the hidden layer (for example, the letters C, H, I, L, O).

Test your network, save it as a JSON file, and export the PDF containing the role cards. (For this, the option "Calculate bias automatically" must be enabled.)