Fuzzy models face constraints in adapting to dynamically changing environments. These models are constructed and trained assuming that underlying data-generating processes remain static over time.

Limited ability to incorporate new information post-training leads to the necessity of retraining the model with updated data. The stability-plasticity dilemma describes the phenomenon where the system's fixed learning capacity results in degraded computing performance and the loss of previously learned knowledge when accommodating new information.

Organizes the extracted knowledge using a set of easily interpretable Mamdani-type IF-THEN fuzzy rules

Dynamically constructs new rules from novel training data and strategically prunes outdated rules

Introduces a localized learning procedure designed for adapting the fuzzy set parameters, balancing interpretability and modeling accuracy

Online Identification of a Nonlinear Dynamic System With Nonvarying Characteristics

Online Identification of a Nonlinear Dynamic System With Time-Varying Characteristics

Lack of a fully online self-reorganizing neurofuzzy system.

Inadequate adaptation of existing neurofuzzy models to handle time-variant data sets

Insufficiency in exploring the neural aspects of fuzzy-associative learning

Absence of a comprehensive understanding of the BCM theory of meta-plasticity for associative learning in the context of self-organizing neurofuzzy systems.

Developed a self-reorganizing fuzzy associative machine (SeroFAM) based on the BCM theory for computational learning.

Applied discrete recursive BCM to facilitate online learning and update of premise nodes and fuzzy rule potentials.

Implemented a parameterization strategy using the half-life concept, influencing various aspects of SeroFAM's functionality.

Employed a decoupled online clustering function for portable membership definitions and potential interoperability between fuzzy systems.

time-variant S&P-500 market index over a period of about 60 years -> reorganize its fuzzy associations once in every 2–3 years from 1980 to 2009

under time-invariant conditions, produce a fuzzy approximation for a nonlinear plant system, achieve a reasonable representation of the plant using simple fuzzy approximates

Subjectivity and inconsistency in the design of neural fuzzy systems by human experts.

Issues related to self-organization of numerical training data, including inconsistent rulebases, the need for prior knowledge like the number of clusters, heuristically designed knowledge acquisition methodologies, and the stability-plasticity tradeoff of the system.

The need for a more efficient, automated, and data-driven approach for the formulation of the fuzzy rulebase and clustering techniques in the design of neural fuzzy systems.

Addressing the stability-plasticity tradeoff by incorporating new clusters while refining existing ones in the input-output space, thus maintaining a balance between past and future knowledge.

Handling the issue of conflicting rulebases by retaining the most significant rule, ensuring the consistency of the SaFIN model's rulebase in describing the application environment accurately.

Proposing a knowledge acquisition methodology inspired by human behavioral category learning processes, leading to the development of a new single-pass fuzzy partitioning technique known as CLIP.