Deep learning by Andrew Ng---DNN

When should do we use fine-tuning?

It is typically used only if you have a large labeled training set; In this setting, fine-tuning can significantly improve the performance of your classifier. However, if you had a large unlabeled dataset (for unsupervised feature learning/pre-training) and only a relatively smal L labeled training Set, then fine-tuning was significantly less likely to help.

Stacked Autoencoders (Training):

Equivalent to capturing the characteristics of an input set with multiple Autoencoder. The first autoencoder captures the characteristics of the dataset and gets the feature matrix1 (hidden Layer weight). Then the feature matrix1 and the input set feedforward processed activation as input to capture higher-level features matrix2 (hidden Layer weight). And then repeat, then the last obtained feature activation as input set input into Softmax classifier (or other classifier) training. ( note that the feature matrix obtained after training is not passed directly to the next autoencoder, but rather the input set is passed to the next autoencoder with the activation of this input set, using the Feedforward method. The output will be passed to the next Autoencode).



Then the entire network after training, the various steps to get the characteristics of the matrix and the parameters of the classifier to synthesize a new network.

Fine-tuning:

(not to be continued)

Deep learning by Andrew Ng---DNN