Backward or reconstruction phase of RBM

In the reconstruction phase, the data from the hidden layer is passed back to the visible layer. The hidden layer vector of probabilities h0 is multiplied by the transpose of the weight matrix W and added to a visible layer bias vb, which is then passed through a sigmoid function to generate a reconstructed input vector prob_v1.

A sample input vector is created using the reconstructed input vector, which is then multiplied by the weight matrix W and added to the hidden bias vector hb to generate an updated hidden vector of probabilities h1.

This is also called Gibbs sampling. In some scenarios, the sample input vector is not generated and the reconstructed input vector prob_v1 is directly used to update the hi

Getting ready

This section provides the requirements for image reconstruction using the input probability vector.

• mnist data is loaded in the environment
• The RBM model is trained using the recipe Training a Restricted Boltzmann machine

How to do it...