This project will consist on the theoretical and experimental analysis of Restricted Boltzmann Machines where the matrix of weights is defined as a function of a set of adjustable patterns, derivation of the learning equations, implementation of the model and experiments in both artificial and real cases. Due to its probabilistic nature and its nice mathematical formulation, RBM differs from many other machine learning algorithms. In particular, once the model is trained, you can obtain additional information not given by the other algorithms, such as the probabilities of new instances. The objective is to explode the real potential as probabilistic model of a RBM. We will investigate a new algorithm called RAPID. It has been implemented and it has been analysed to detect its deficiencies. The algorithm presents some issues which we will try to solve implementing some modifications. Then, we will compare the results with common algorithms as CD. The experiments demonstrate that this new algorithm could be an interesting procedure to reduce the computational cost of training some models.