可以使用的一个句式
我们这篇文章中的方法可以在目前的两种方法中取得一个平衡。这篇文章来自 hypernetwork, 2017 google team
In this paper, we view convolutional networks and recurrent networks as two ends of a spectrum.
On one end, recurrent networks can be seen as imposing weight-sharing across layers, which makes them inflexible and difficult to learn due to vanishing gradient. On the other end, convolutional networks enjoy the flexibility of not having weight-sharing, at the expense of having redundant parameters when the networks are deep. Hypernetworks can be seen as a form of relaxed weightsharing, and therefore strikes a balance between the two ends. See Appendix A.2 for conceptual diagrams of Static and Dynamic Hypernetworks.