Skip to content
conftrace_

depth-scaled initialization

1 paper

Explore in graph

Also known as

DS-INIT

Co-occurring keywords · papers mentioning both

merged attention (1) encoder-decoder attention (1) deep transformer (1) layer normalization (1) residual connection (1) gradient vanishing (1) model convergence (1) deep neural network (1) machine translation (1)

Papers

Improving Deep Transformer with Depth-Scaled Initialization and Merged Attention EMNLP 2019