Domain Adversarial Neural Networks

All machine learning algorithm assumes that the train and test data comes from the same distribution. For example, when we train a classifier on handwritten digits of the USPS dataset, it performs much poorer on the MNIST dataset. In DANN, this is solved adversarially by using the a discriminator network (identifies data from source and target domain) that forces the feature extractor to produce similar features for both the domains.

Note: The DANN doesn’t see the labels of the target data. Yet it increases the cross-generalization (adaptation to the target data) performance by 30%. Find the code here.