These are wrappers for Stateful, TemporalStateful and SpatialStateful interfaces which reverse gradients on the states during the back-propagation. This can be used for adversarial training of classifiers/decoders by improving the classifier itself, but making the rest of the network to prevent the classifier to do a good job.
These are wrappers for
Stateful
,TemporalStateful
andSpatialStateful
interfaces which reverse gradients on the states during the back-propagation. This can be used for adversarial training of classifiers/decoders by improving the classifier itself, but making the rest of the network to prevent the classifier to do a good job.