Doubt_regarding_model_structure

[akshayaCap]

Please correct me if I am wrong: Stacked Denoising Autoencoder needs to be trained layerwise upto my knowledge. I cannot see the code written in this fashion. Please correct me if I am missing on something.

[jonlu0602]

At first, layer-wise training was useful for denoising autoencoder in small dataset. However, we can improve our performance by fitting a lot of data or RNN to predict our denoising spectrogram. That's why I didn't train in layer-wise way.

references:
https://hal.inria.fr/hal-01163493/document
https://www.citi.sinica.edu.tw/papers/yu.tsao/6258-F.pdf

[akshayaCap]

At first, layer-wise training was useful for denoising autoencoder in small dataset. However, we can improve our performance by fitting a lot of data or RNN to predict our denoising spectrogram. That's why I didn't train in layer-wise way.

references:
https://hal.inria.fr/hal-01163493/document
https://www.citi.sinica.edu.tw/papers/yu.tsao/6258-F.pdf

Thank-you for clarification. I went through the code and could not find this method of pre-processing included. Do you have any other repository where you have implemented this?