Interval Type-2 Fuzzy Restricted Boltzmann Machine

Restricted Boltzmann machine (RBM) is an energy-based artificial neural network (ANN), applied in several applications like image processing, topic modeling, classification, regression, and pattern recognition. The fuzzy version of RBM is a new approach in this field, with parameters considered as fuzzy numbers. In this article, a fuzzy RBM is extended through interval type-2 membership functions, named the interval type-2 fuzzy RBM (IT2FRBM). The additional uncertainties in the structures of the membership functions are embedded in this model. This is formulated as a maximum likelihood problem which allows the parameters of the type-2 fuzzy numbers to be learned. The capabilities of this proposed approach as a discriminative or generative model are assessed. The robustness of this method against noise is analyzed. The results indicate that this IT2FRBM outperforms RBM and its different fuzzy versions.

Automated summary

The article introduces a new type of fuzzy Boltzmann machine, interval type-2 fuzzy RBM (IT2FRBM), which utilizes interval type-2 fuzzy membership functions to handle additional uncertainties and has the capability to learn parameters of fuzzy numbers. The research findings suggest that IT2FRBM outperforms traditional RBM and other fuzzy versions in both discriminative and generative modeling.