A brief survey on deep belief networks and introducing a new object oriented MATLAB toolbox (DeeBNet)

Nowadays this is very popular to use deep architectures in machine learning. Deep Belief Networks (DBNs) are deep architectures that use stack of Restricted Boltzmann Machines (RBM) to create a powerful generative model using training data. DBNs have many ability like feature extraction and classification that are used in many application like image processing, speech processing and etc. The paper provides a survey of the relevant literatures on DBNs and introduces a new object oriented MATLAB toolbox with most of DBN's abilities. According to the results on MNIST (image dataset) and ISOLET (speech dataset), the toolbox can extract useful features with acceptable discrimination between them without using label information. Also on both datasets, the obtained classification errors are comparable to the state of the arts literatures on them. In addition the toolbox can be used in other applications like generating data from trained model, reconstructing data and reducing noise. The toolbox is open source software and freely available on the website http://ceit.aut.ac.ir/~keyvanrad/ .