Incremental Cluster Validity Index-Guided Online Learning for Performance and Robustness to Presentation Order

Area of Science:

• Machine Learning
• Data Mining
• Artificial Intelligence

Background:

• Lifelong learning systems require intelligent decision-making for streaming data where samples are processed and discarded.
• The order of incoming data significantly impacts the performance of incremental learning algorithms.
• Incremental cluster validity indices (iCVIs) have emerged as valuable tools for cluster quality monitoring and have been integrated into streaming clustering methods.

Purpose of the Study:

• To introduce the first adaptive resonance theory (ART)-based model utilizing iCVIs for unsupervised and semi-supervised online learning.
• To demonstrate how iCVIs can regulate ART vigilance through an iCVI-based match tracking mechanism.
• To enhance the accuracy and robustness of online learning models against data ordering effects.

Main Methods:

• Development of iCVI-TopoARTMAP, an ART-based model integrating an online iCVI module into a topological ART predictive mapping (TopoARTMAP) architecture.
• Implementation of an iCVI-based match tracking mechanism to regulate ART vigilance.
• Application of iCVI-driven postprocessing heuristics at the end of each learning iteration.

Main Results:

• The iCVI-TopoARTMAP model demonstrated improved accuracy and robustness to data presentation order compared to existing methods.
• The online iCVI module effectively assigned input samples to clusters in each iteration.
• Experimental evaluation on synthetic and real-world datasets confirmed the model's performance and robustness.

Conclusions:

• The proposed iCVI-TopoARTMAP model offers a significant advancement in online learning for streaming data applications.
• Integrating iCVI with ART models provides a robust framework for unsupervised and semi-supervised learning, overcoming limitations of data ordering.
• The model retains beneficial ART properties like stability and immunity to catastrophic forgetting.