Self-boosting first-order autonomous learning neuro-fuzzy systems

Abstract

In this paper, a detailed mathematical analysis of the optimality of the premise and consequent parts of the recently introduced first-order Autonomous Learning Multi-Model (ALMMo) neuro-fuzzy system is conducted. A novel self-boosting algorithm for structure-and parameter-optimization is, then, introduced to the ALMMo, which results in the self-boosting ALMMo (SBALMMo) neuro-fuzzy system. By minimizing the objective functions with the previously collected data, the SBALMMo is able to optimize its system structure and parameters in few iterations. Numerical examples based benchmark datasets and real-world problems demonstrate the effectiveness and validity of the SBALMMo, and show the strong potential of the proposed approach for real applications.

Particle Swarm Optimized Autonomous Learning Fuzzy System

“…However, the system performance will deteriorate significantly if η o is set too large since the system forgets the learned knowledge from historical data rather rapidly. The influence of Ω o and η o has been analyzed and verified through numerical examples in [7], [28]. The recommended values of Ω o and η o are 10 and 0.1, respectively.…”

“…However, this is, again, a"one pass" approach and it lacks a theoretical proof of optimality. In [28], the optimality of EISs is systematically studied, and two algorithms are introduced for optimizing the antecedent and consequent parts, respectively. Nonetheless, the prediction accuracy is only improved marginally on certain problems after system optimization.…”

“…Nonetheless, the prediction accuracy is only improved marginally on certain problems after system optimization. This is mainly because the antecedent and consequent parts of the system are optimized separately, based on the use of different criteria [28]. The overall performance of an EIS is decided by both antecedent and consequent parts.…”

Optimize TSK Fuzzy Systems for Regression Problems: Minibatch Gradient Descent With Regularization, DropRule, and AdaBound (MBGD-RDA)

“…• Layer 2: The firing level of each rule Rule r is computed, by (3). • Layer 3: The normalized firing levels of the rules are computed, by (5). • Layer 4: Each normalized firing level is multiplied by its corresponding rule consequent.…”

“…There are generally three different strategies for optimizing a TSK fuzzy system in supervised regression 1 : 1) Evolutionary algorithms [6], in which each set of the parameters of the antecedent membership functions (MFs) and the consequents are encoded as an individual in a population, and genetic operators, such as selection, 1 Some novel approaches for optimizing evolving fuzzy systems have also been proposed recently [4], [5]; however, they are not the focus of this paper, so their details are not included. crossover, mutation, and reproduction, are used to produce the next generation.…”