| Summary: | 碩士 === 國立成功大學 === 工業管理學系 === 88 === Training examples are always the key point to affect the performance of supervised learning system, especially in example size and information sufficiency that examples can provide to learning systems. This paper uses back-propagation neural network as knowledge acquisition tool. A FMS dynamic scheduling system (4machines,4 parts, and 8 buffers) is built to generating examples. And we introduce the concept of "function virtual population" to improve classification rate of the neural network. Using this method, we not only can improve drastically the performance of this learning system, but also find suitable example size.
|
|---|
