Modular Learning in Neural Networks: A Modularized Approach to Neural Network Classification

By Tomas Hrycej

Provides evidence that modular learning is helpful in improving learning performance. Numerous approaches were tested including decomposition of learning into modules using various learning types (supervised and unsupervised learning); decomposition of the mapping to be represented (linear and nonlinear parts); decomposition of the neural network to minimize harmful interaction during learning; decomposition of the application task into subtasks that are learned separately; and decomposition into a knowledge-based part and a learning part. These methods were tested on two ''benchmark'' cases--a medical classification problem (7,200 cases of thyroid disorder) and a handwritten digits classification problem (20,000 cases).

• Format: Hardcover
• Language: English
• ISBN: 0471571547
• ISBN13: 9780471571544
• Release Date: September 1992
• Publisher: Wiley-Interscience
• Length: 256 Pages
• Weight: 1.20 lbs.
• Dimensions: 9.6" x 0.9" x 6.4"