讨论了一种基于前馈型极值学习机的深度神经网络隐层参数整定问题。
Tuning the hidden layers parameters for a deep neural network with the type of feed forward based Extreme learning machine is discussed.
使用基于ELM的深度信念网络,您可以通过将输入映射到另一个空间,并根据需要使用另一个维度来重建输入信号。
Using Deep belief networks based ELM you can reconstruct the inputs by mapping them into another space with another dimensions as you desire.
然后,您将有可能减少或增加使用堆叠自动编码器输入的维数,这意味着您可以完全调整隐藏层的所有参数,而不仅仅是输出权重β,如通在基本ELM理论中提到的单隐藏层前馈神经网络那样容易。
And later you will have the possibility to decrease or increase the dimensionality of your inputs using stacked auto-encoders, and this is means that this time you can tune all the parameters of the hidden layers entirely not only the output weights beta as mentioned in basic ELM theories for single hidden layer feedforward neural network.
使用堆叠的自动编码器调整隐藏层所有参数可以保护最佳的网络泛化和近似。
Tuning all the parameters of the hidden layers using stacked auto encoders allows you to guard the best networks generalization and approximation.
在该代码中,我们向您展示了一个基于ELM的DBN演示版本,它应用于OCR字母数据集MINST的一小部分,如果您对整个代码感兴趣,请随时与我联系。
In this code we represent to you a demo version of DBNs base ELM applied on small part of MINST of OCR letters dataset, if you are interested about the entire code don’t hesitate to contact me.
作者邮箱:berghouttarek@gmail.com
完整代码下载地址:
http://page2.dfpan.com/fs/elcj4221b2918600d92/
更多精彩文章请关注微信号: