Related Experiment Videos
Fuzzy ARTMAP: A neural network architecture for incremental supervised learning of analog multidimensional maps
G A Carpenter1, S Grossberg, N Markuzon
1Center for Adaptive Syst., Boston Univ., MA.
IEEE Transactions on Neural Networks
|January 1, 1992
Summary
A novel fuzzy ARTMAP neural network architecture enables incremental supervised learning for recognition and mapping. This fuzzy logic and adaptive resonance theory synthesis demonstrates strong performance across various simulations.
Area of Science:
- Artificial Intelligence
- Machine Learning
- Computational Neuroscience
Background:
- Traditional neural networks face challenges in incremental learning and fuzzy feature representation.
- Adaptive Resonance Theory (ART) and fuzzy logic offer complementary strengths for pattern recognition.
Purpose of the Study:
- Introduce fuzzy ARTMAP, a novel neural network architecture.
- Synthesize fuzzy logic and ART for enhanced incremental supervised learning.
- Evaluate fuzzy ARTMAP's performance against benchmark systems.
Main Methods:
- Developed a neural network architecture integrating fuzzy logic and ART principles.
- Utilized fuzzy subsethood computations within the ART framework for category learning.
- Conducted simulations on diverse tasks including geometric recognition, function approximation, and character recognition.
Main Results:
- Fuzzy ARTMAP demonstrated effective incremental supervised learning capabilities.
- The system showed competitive or superior performance compared to backpropagation and genetic algorithms.
- Evaluated against NGE and FMMC systems, highlighting fuzzy ARTMAP's efficacy.
Conclusions:
- Fuzzy ARTMAP offers a robust framework for incremental supervised learning.
- The integration of fuzzy logic and ART provides significant advantages in recognition and mapping.
- The architecture shows promise for complex pattern recognition tasks with evolving data.