A hybrid ART-GRNN online learning neural network with a epsilon -insensitive loss function

Area of Science:

• Artificial Intelligence
• Machine Learning
• Neural Networks

Background:

• Generalized Regression Neural Networks (GRNN) are effective but can lack online learning capabilities.
• Adaptive Resonance Theory (ART) offers online learning but may require modifications for certain tasks.
• Hybrid models can combine the strengths of different neural network architectures.

Purpose of the Study:

• Introduce a novel hybrid neural network model, Generalized Adaptive Resonance Theory (GART).
• Enhance GRNN by integrating online learning properties from ART.
• Evaluate GART's effectiveness across diverse machine learning tasks.

Main Methods:

• Developed GART by combining Modified Gaussian Adaptive Resonance Theory (MGA) with GRNN.
• Conducted empirical studies on classification, regression, and time series prediction datasets.
• Compared GART's performance against established methods like Online Sequential Extreme Learning Machine (OSELM) and sequential learning Radial Basis Function (RBF) networks.

Main Results:

• GART demonstrated robust performance in classification tasks.
• GART achieved competitive results in regression analysis.
• The model showed effectiveness in time series prediction.

Conclusions:

• GART successfully integrates online learning into a GRNN framework.
• The hybrid model offers a viable alternative to existing sequential learning neural networks.
• GART presents a promising approach for various machine learning applications requiring continuous learning.