Linear Transformations for multilayer-perceptron neural networks, described in paper
Pushing stochastic gradient towards second-order methods–backpropagation learning with transformations in nonlinearities T Vatanen, T Raiko, H Valpola, Y LeCun Neural Information Processing, 442-449
available online at http://arxiv.org/pdf/1301.3476.pdf