| 摘要 |
A process for tuning the interconnection strengths in artificial neural networks (ANNs) and similar machines. The process first generates a series of ANNs by simulating the steepest descent process on an error function, each time with one or more of the available observations withheld in order to measure the predictive performance of the ANN. Finally, the process is repeated without withholding any observations, using the previously obtained information about predictive performance to identify the best performing of the ANNs generated in the final process.
|
