A new approach of a possibility function based neural network


The paper presents a new type of fuzzy neural network, entitled Possibility Function based Neural Network (PFBNN). Its advantages consist in that it not only can perform as a standard neural network, but can also accept a group of possibility functions as input. The PFBNN discussed in this paper has novel structures, consisting in two stages: the first stage of the network is a fuzzy based and it has two parts: a Parameter Computing Network (PCN), followed by a Converting Layer (CL); the second stage of the network is a standard backpropagation based neural network (BPNN). The PCN in a possibility function based network can also be used to predict functions. The CL is used to convert the possibility function to a value. This layer is necessary for data classification. The network can still function as a classifier using only the PCN and the CL or only the CL. Using only the PCN one can perform a transformation from one group of possibility functions to another.

Publication Title

Advances in Intelligent Systems and Computing