如何在深度学习网络中强化前向传播信号？

Question

I've asked a question earlier at Matconvnet output of deep network's marix is uniform valued instead of varying values? As I debugged the deep network for density estimation, I realized the signal towards the output dies out/fades. How can I reinforce the signal so that it doesn't become flat- and the peak and base values keep a high amplitude?

Answer 1

在 LRN 归一化中使用较低的 alpha 可以增强输出。