Polish documents explaining how this network works can be found in Docs\ directory.
The solution consists of three projects:
- NetIn
- NeuronNetwork
- NeuronNetwork.MatrixInput
Whole solution tries to teach the network to be able to recognize if given matrices are "correct". Every matrix is binary and 3x3. In our basic problem we have 5 such matrices. Each of these matrices corresponds to a single digit. The number of "ones" describes which number is the matrix.
So for example this matrix corresponds to number 6:
| 1 | 1 | 0 |
| 0 | 1 | 0 |
| 1 | 1 | 1 |
In this project I'm trying to find a number 11027 but you can set it to anything else.
This project is a console application used for running the network. It references the other two projects and the main aspect of this project is to create the neuron network and test when it will learn the given problem.
This project is a library project with most of the Neuron Network standard logic (Perceptrons and connections between them)
This project is a library which references to NeuronNetwork. In it you can find a simple class Matrix
which is used for the single digit representations (and it gives you a possibility to create some random matrices).
Also there is a class Called MatrixInputNetwork which gives all network nodes a proper order of making calculations
and backwards propagations.
Bartosz Kalinowski s11027