c++ implementation of multi-layer feed forward neural networks with back propagation algorithm. Optimization + features to be added soon.

main.cpp contains an example of XOR function learning and a more advanced application case: digit recognition.

Console output for XOR network :

Number of layers : 3
Number of input neurons : 2
Number of hidden neurons : 4
Number of output neurons : 1

Test at iteration 3475 :
XOR(0,0) = 0
XOR(0,1) = 1
XOR(1,0) = 1
XOR(1,1) = 0

Visualization of non-linear discrimination for XOR :

Samples used can be found in samples/*.txt. Learning and test variables are already processed from original histogram pool through normalization and PCA projection.

X* : experiences * variables
T* : targets

Dicrimination result of a neural network trained with 967 experiences in 55 dimensions tested with 967 unlearned inputs :

Stopping learning at iteration : 1000
Correct : 902
Incorrect : 65
Ratio : 93.2782

Note that performances are strongly dependent of representation choices in variable space.

• http://www.cs.bham.ac.uk/~jxb/NN/l7.pdf
• http://www.di.unito.it/~cancelli/retineu06_07/FNN.pdf
• ESEO Data Mining Course - M. Feuilloy