Fuzzy Neural Networks (FNNs) Training Algorithm With Dropout via Its Equivalent Fully Connected Fuzzy Inference Systems (F-CONFIS)

Fuzzy neural network (FNN) often suffers from overfitting problem, especially when FNN has large number of parameters. In the FNN system, there are two types of adjustable parameters, one is control parameters, and the other is link weights of consequent part. To improve convergent rate, Dropout technique is first adopted for Fuzzy neural network. A new training algorithm with dropout technique for FNN is proposed via its equivalent F-CONFIS. Illustrative examples are provided for checking the validity of the proposed method. Simulation attained satisfactory results. Proposed method for Fuzzy neural network via F-CONFIS has its rising values in all practical applications, such as system identification, expert System and image information processing system …, etc.