论文部分内容阅读
针对在建立定量构效关系(QSAR)模型中,单个人工神经网络模型难以确定参数,容易产生“过拟合”;一般神经网络集成模型虽然建立过程简单,但由于个体差异度小而导致泛化能力相对单个神经网络没有明显改善等问题,提出了一种基于随机梯度法的选择性神经网络二次集成方法。在建立除草剂(苯乙酰胺类化合物)的QSAR模型的实验研究中表明,该方法设计过程简单,能够以较小的运算代价明显地提高了模型的泛化能力,是建立QSAR模型的一个有效方法。
In the establishment of quantitative QSAR model, single artificial neural network model is difficult to determine the parameters, prone to “over-fitting”; general neural network integrated model although the establishment process is simple, but due to small differences in individual and lead to generalization The capability is not improved significantly compared to a single neural network and so on. A new method of quadratic integration based on stochastic gradient is proposed. The experimental study of establishing QSAR model of herbicide (phenylacetamides) shows that this method has a simple design process and can significantly improve the generalization ability of the model with a little computational cost. It is an effective way to establish the QSAR model method.