lstm是怎么解决梯度消失问题的

https://www.quora.com/How-does-LSTM-help-prevent-the-vanishing-and-exploding-gradient-problem-in-a-recurrent-neural-network#:~:text=The%20vanishing%20(and%20exploding)%20gradient%20problem%20is%20caused%20by%20the,controlled%20by%20the%20forget%20gate.&text=Compared%20with%20LSTM%2C%20it%20is,be%20stacked%20into%20deep%20models.

这篇文章，从直觉的角度讲解了下，我觉得比较容易理解。

 

the LSTM architecture is resistant to exploding and vanishing gradients, although mathematically both phenomena are still possible.

LSTM这个结构其实并没有彻底解决梯度的消失或爆炸问题，只是有所缓解。因为除了cell单元的更新是改为加法外，其他的路径上仍然是做乘法，所以在其他路径上仍然存在梯度的消失或爆炸问题。