Noise-tolerant neural networks for pattern classification and function extrapolation
To study and improve the effectiveness and potential of neural networks in pattern classification and function extrapolation under noise environment, we propose two noise reduction algorithms based on training samples to enhance the capability of neural networks in noise environment and construct a...
| Main Author: | |
|---|---|
| Other Authors: | |
| Format: | Thesis |
| Published: |
2008
|
| Subjects: | |
| Online Access: | http://hdl.handle.net/10356/4895 |
| Summary: | To study and improve the effectiveness and potential of neural networks in pattern classification and function extrapolation under noise environment, we propose two noise reduction algorithms based on training samples to enhance the capability of neural networks in noise environment and construct a new network structure to realize noise-tolerant short-term and long-term forecasting. The detrimental effect of overlapping data and noise in neural networks that cause over-learning as a result to substantially deteriorate neural networks' performance has also been studied in this thesis. |
|---|