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).
Abstract (translated by Google)
URL
https://arxiv.org/abs/1904.11816