Adaptive behavior with fixed weights in RNN: an overview

In this paper we review recent results on the adaptive behavior attained with fixed-weight recurrent neural networks (meta-learning). We argue that such behavior is a natural consequence of prior training.