Under review as a conference paper at ICLR 2019 softmax Transpose Updated Memory Memory embedded input linear concat Attention Weights concat

Subfields: Model Architectures