对于多输出神经网络，有种特别简单的办法求平均值，比如说：
If there is a one-many relationship between the inputs and targets in the training data, then it is not possible for any mapping of the form to perform perfectly. It is straightforward to show that if a probability density P(Y|X) describes the data, then the minimum of error measure is attained by the map taking X to the average target.
Any given network might or not be able to approximate this mapping well, but when trained as well as possible it will form its best possible approximation to this mean.
Of course, there are a lot of P(Y|X), which will influence the results.
【免责声明】本文仅代表作者个人观点，与IT09数码网无关。其原创性以及文中陈述文字和内容未经本站证实，对本文以及其中全部或者部分内容、文字的真实性、完整性、及时性本站不作任何保证或承诺，请读者仅作参考，并请自行核实相关内容。您若对该稿件内容有任何疑问或质疑，请联系本网将迅速给您回应并做处理。