与 《Knowing When to Look: Adaptive Attention via A Visual Sentinel for Image Captioning》...
![240](https://cdn2.jianshu.io/assets/default_avatar/12-aeeea4bedf10f2a12c0d50d626951489.jpg?imageMogr2/auto-orient/strip|imageView2/1/w/240/h/240)
IP属地:湖北
与 《Knowing When to Look: Adaptive Attention via A Visual Sentinel for Image Captioning》...
Motivation plain的RNN存在梯度消失与爆炸问题,而LSTM/GRU虽然一定程度解决了问题,但是由于采用tanh或者sigmoid作为这些饱和的激活函数,堆叠多...