BriefGPT.xyz
Mar, 2021
UnICORNN: 用于学习非常长时间依赖关系的循环模型
UnICORNN: A recurrent model for learning very long time dependencies
HTML
PDF
T. Konstantin Rusch, Siddhartha Mishra
TL;DR
本文提出了一种基于哈密顿系统的离散化的循环神经网络架构，解决长时依赖序列输入处理的梯度消失和爆炸问题，实验表明该方法在各种学习任务中提供了最先进的性能。
Abstract
The design of
recurrent neural networks
(RNNs) to accurately process sequential inputs with
long-time dependencies
is very challenging on account of the
→