Decentralized Nonstationary Fuzzy Neural Network with Meta-Learning-Net

The nonstationary fuzzy neural network (NFNN) has proven to be an effective and interpretable tool in machine learning, capable of addressing uncertainty problems similarly to type-2 fuzzy neural networks, while offering reduced computational complexity. However, the update of disturbance parameters in an NFNN is restricted due to the necessity of maintaining a regular membership function, which limits its learning capability. To address this limitation, we propose a Decentralized NFNN (DNFNN) that overcomes the obstacles in the backward update process and enhances the efficiency of large-scale optimization. Additionally, we demonstrate improved computational efficiency and establish the linear convergence of the proposed decentralized algorithm. By integrating a meta-learning network, we further enhance the output strategy of the NFNN, enabling it to adaptively determine the contribution of individual sub-networks. Experimental results on various UCI datasets, spanning multiple domains and exhibiting diverse dimensions and sizes, show that the DNFNN outperforms existing methods in terms of classification accuracy, robustness and practicality.