Deep Belief Networks is a type of neural network referred to as the process of making the neural network generates the high probability by training the weights in between the neurons. It is a generative graphical model comprised of many hidden layers connecting layers but not between the units within each layer.

DBN is accustomed to being either a supervised or unsupervised model. It belongs to greedy learning algorithms and consists of multiple layers of neurons categorized into visible and hidden neurons. The visible neurons are used to accept input, and the hidden neurons extract features.

This pre-trained algorithm learns all the top-down perspectives and most important generative weights using a layer-by-layer approach. DBN is constructed by one dimension of the data vector is represented by each neuron of the bottom layer, the bottom layer denotes the data vectors, the layers of DBN are connected layer by layer.

The process of training DBN is done layer by layer. In each layer, the data vector is used to infer the hidden layer, and this hidden layer is treated as the data vector of the next layer (higher layer).

The most popular applications of Deep Belief Network are image classification, generating and recognizing images, image compression, video generation, motion capture, and speech recognition. Recent advances of deep belief networks are intrusion detection, medical image and data analysis, posture estimation, intelligent fault detection, fault prediction, prediction of reservoir landslide displacement, and more.

   • Deep Belief Networks (DBNs) are probabilistic generative models that extract a deep hierarchical representation of the training data.

   • DBN employs stacking of RBM models with an efficient layer-by-layer greedy learning strategy to initialize the deep network and, in the sequel, fine-tune all weights jointly with the desired outputs.

   • Despite DBN outperforming other models, it is also plagued by several shortcomings involving difficulty in accurately estimating joint probabilities and computational costs associated with training a DBN.

   • In addition, DBNs do not account for the two-dimensional structure of an input image, which may significantly affect their performance and applicability in computer vision and multimedia analysis problems.

   • Owing to large data, the training processes of DBNs are time-consuming and cannot provide satisfactory results in learning speed, modeling accuracy, and robustness, which is mainly caused by dense representation and gradient diffusion.