Add support for feature extracting for federated logistic regression by exploiting DNN

[yankang18]

For now, the federated logistic regression (LR) algorithm is only using structural data (i.e., tabular data). This limits the application of LR. We may add support for automatic feature engineering to LR for dealing with various types of inputs such as text and images.

Neural networks such as RNN, CNN and autoencoders are widely used for learning features from text and images. Therefore, we may add these neural networks as local models for parties to extract features and then feed extracted features to LR.

This feature is recommended for FATE 0.3v

[computereasy]

@yankang18 @dylan-fan Hey there, just a quick question regarding the progress of RNN/CNN support. How is it going so far? Can we expect to use DNN for parties to extract features like images? Thank you very much.