Think Again Networks and the Delta Loss

Abstract
This short paper introduces an abstraction called Think Again Networks (ThinkNet) which can be applied to any state-dependent function (such as a recurrent neural network).
View on arXivComments on this paper