Related Experiment Videos
A selective learning method to improve the generalization of multilayer feedforward neural networks
1Universidad Carlos III de Madrid, Avenida de la Universidad, 30, 28911 Leganés, Madrid, Spain.
International Journal of Neural Systems
|November 25, 2003
Summary
This study introduces a novel lazy learning method for neural networks that dynamically selects training data. This approach improves generalization performance by focusing on relevant patterns for new predictions.
Area of Science:
- Artificial Intelligence
- Machine Learning
- Neural Networks
Background:
- Multilayer feedforward neural networks with backpropagation are widely used.
- Generalization performance heavily relies on training data quality, with redundant or irrelevant patterns posing challenges.
Purpose of the Study:
- To present a novel learning method for automatic selection of training patterns.
- To improve generalization abilities of neural networks by adapting to new samples.
Main Methods:
- A lazy learning strategy is employed, building approximations centered around the novel sample.
- The method dynamically selects training patterns most appropriate for predicting new data points.
Main Results:
- The proposed method was applied to artificial approximation problems and a real time series prediction task.
- Compared to standard backpropagation, the new method demonstrated superior generalization abilities.
Conclusions:
- The developed learning method enhances neural network generalization by intelligently selecting training data.
- This approach offers a more effective way to handle training data, leading to better predictive performance.