Comparison of a Neural Network Based on Fuzzy Flip-Flops and an MLP Robustness in Function Approximation

In this paper two types of neural networks, namely the “traditional” tansig based neural networks and the multilayer perceptrons based on fuzzy flip-flops (F3NN) trained by the Bacterial Memetic Algorithm with Modified Operator Execution Order (BMAM) are tested and compared on their robustness to test functions outliers. The robust design of the F3NN is presented, and the best suitable fuzzy neuron type is emphasized. As our major motivation in these investigations was to construct a technology for the creation of real hardware MLPs and for this reason the fuzzy flip-flop based F3NNs obviously offered much simpler and cheaper possibility for hardware implementation compared to a relatively complicated tansig type neural network.