Neural network using back propagation framework for training, momentum strategy included. The gradient descent algorithm is stochastic gradient descent. This network supports multi hidden layers. The main function is an example of simple antoencoder for sparse vectors.