Robust Fuzzy Neural Network With an Adaptive Inference Engine.

Area of Science:

• Artificial Intelligence
• Machine Learning
• Fuzzy Systems

Background:

• Fuzzy neural networks (FNNs) excel at handling data uncertainty but face generalization and dimensionality challenges.
• Deep neural networks (DNNs) process high-dimensional data but have limited capacity for uncertainty.
• Existing deep learning methods for robustness are often time-consuming or perform poorly.

Purpose of the Study:

• To propose a robust fuzzy neural network (RFNN) that overcomes the limitations of traditional FNNs and DNNs.
• To develop an adaptive inference engine capable of handling high-dimensional data with high uncertainty.
• To enhance the reasoning ability of fuzzy rules for complex inputs.

Main Methods:

• Introduced a robust fuzzy neural network (RFNN) with an adaptive inference engine.
• Developed a method for adaptive learning of firing strengths and processing uncertainty in membership function values.
• Utilized neural network structures in the consequent layer to improve reasoning for complex inputs.

Main Results:

• The proposed RFNN demonstrates state-of-the-art accuracy on various datasets, even with high levels of uncertainty.
• The adaptive inference engine effectively handles samples with high uncertainty and dimensionality.
• Learned fuzzy sets automatically cover the input space well, enhancing network performance.

Conclusions:

• The RFNN offers a robust solution for problems involving high-dimensional data and significant uncertainty.
• This approach improves upon traditional FNNs and DNNs by integrating adaptive learning and enhanced reasoning.
• The developed RFNN provides superior accuracy and robustness in complex machine learning tasks.