Related Experiment Videos
New training strategies for constructive neural networks with application to regression problems.
1Department of Electrical and Computer Engineering, Concordia University, 1455 De Maisonneuve Blvd West, Montreal, Que. H3G 1M8, Canada. liying@ece.concordia.ca
Summary
This study introduces novel incremental constructive training methods for one-hidden-layer feedforward neural networks (OHL-FNNs). These techniques improve training efficiency and generalization performance in regression problems.
Area of Science:
- Artificial Intelligence
- Machine Learning
- Neural Networks
Background:
- Feedforward Neural Networks (FNNs) are widely used for regression problems.
- One-Hidden-Layer Feedforward Neural Networks (OHL-FNNs) offer a simpler architecture with significant learning capabilities.
Purpose of the Study:
- To develop and evaluate novel incremental constructive training schemes for OHL-FNNs.
- To enhance training efficiency and generalization performance in regression tasks.
Main Methods:
- Proposed incremental constructive training with separated input-side and output-side training.
- Introduced a novel error signal scaling technique for improved input-side training efficiency.
- Applied two pruning methods to remove redundant input-side connections.
Main Results:
- The proposed strategies demonstrate potential advantages over existing techniques.
- Numerical simulations validate the effectiveness of the new training and pruning methods.
- Improved generalization performance was observed.
Conclusions:
- The developed incremental constructive training schemes offer an efficient approach for OHL-FNNs in regression.
- Error signal scaling and pruning techniques contribute to better performance and reduced complexity.
- This work advances the application of neural networks in solving regression problems.