| 查看: 205 | 回复: 0 | |||
| 当前主题已经存档。 | |||
[交流]
【求助】急求：神经网络拟合非线性函数的问题！！！
|
|||
|
本人现在用 ”径向基神经网络“ 拟合一条类似于 ”指数曲线“ 的函数，拟合的相对误差小于2.5%，满足了实验的需要。网络的输入和输出向量都是一维的，样本容量为50。训练过程中，输入和输出向量经过log函数和归一化法预处理，并采用了留一法交叉验证。文章修回时，审稿人提出了许多的问题，有些很难回答，希望能得到神经网络高手的指点！我该如何回答这些问题呢？ 1. Certain authors claim an ANN should use at least 5 sample points per connection weight . I believe this is far away from the conditions used by the authors. How was this limitation circumvented by authors? 2. Is there a criteria to choose to normalize data in the interval [0,1] ? 3. Can the authors justify that total numeric range is narrowed and simultaneously the resulting values are uniformly weighed? 4. The use of leave-one-out cross validation implies that the overfitting must be taken into account. This is done in the case of lack of enough input data, given a single sample is excluded each time. In fact, a k-fold-cross validation is much more confident in this respect. Can the authors comment on this point? |
回复此楼» 猜你喜欢
申请科研项目纯靠实力能成功吗？
已经有11人回复
-
40+中年人失业后，还能找到好工作吗？
已经有5人回复
-
极限拉扯，如何抉择
已经有15人回复
-
二审4个多月了还是Awaiting Reviewer Scores
已经有4人回复
-
OpenAI又攻克数学难题了，离AI做有机合成还有多远？
已经有13人回复
-
课题组会帮你通过考核么？
已经有7人回复
-
售一区SCI文章T0P，我:8O.551.O54,科目全,可十急
已经有9人回复
-
售SCI一区文章，我:8O5.5.1.O5.4,科目全,可伽急
已经有10人回复
-
售SCI一区T0P文章，我:8.O55.1.O.54,科目全,可十急
已经有8人回复
-
售SCI一区T0P文章，我:8.O55.1.O.54,科目全,可十急
已经有9人回复
200
