Neural networks for transforming signals

A method for transforming input signals, comprising:

• selecting from a memory a neural network, wherein the neural network includes a set of layers, wherein an activation function of each layer is equivalent to a corresponding iteration of an iterative inference procedure of a generative model, wherein the same set of network parameters is reused by all layers, such that dimensions, ordering, and meaning of parameters are the same for all layers, wherein the iterative inference procedure includes one or combination of a variational inference, a belief propagation, and multiplicative updates, and wherein values of the set of network parameters of at least one layer are different from values of the set of network parameters of at least one other layer;
• and
• transforming the input signals using the neural network to obtain output signals, wherein the steps are performed in a processor operatively connected to the memory.