由于神经网络训练直接由样本集合得到分界平面,在分配出错样本之间来回调整权值,而没有对样本集合本身进行处理,导致延展性较差。比如训练样本A中表现良好的神经网络,在样本B中可能表现较差。而如果同时进行样本A及样本B的训练,则可能面临用时过长的问题。如果对样本集合进行特征提取及化简,则可将任务分配和联合处理。如此,将大幅度提升分界平面的适用范围。 Since the neural network training directly obtains the demarcation plane from the sample set, the weights are adjusted back and forth between the assigned error samples without the sample set itself being processed, resulting in poor scalability. For example, the well-behaved neural network in training sample A may perform poorly in sample B. However, if both sample A and sample B are trained at the same time, the problem of overlong time may be faced. If the sample collection feature extraction and simplification, you can assign tasks and joint processing. In this way, the scope of the demarcation plane will be greatly enhanced.