We propose a novel Convolutional Neural Network (CNN) compression algorithm based on coreset representations of filters. We exploit the redundancies extant in the space of CNN weights and neuronal activations (across samples) in order to obtain compression. Our method requires no retraining, is easy...
通俗的话讲,就是对每个filter关于所有样本和位置的激活值做个均值,作为这个滤波器的绝对重要性,然后对该层所有绝对重要性做一个归一化变成相对的,在构造重构误差时,给每个filter乘一个相对重要性再计算范数。 最后,给出了decomposition的流程,也是利用二分法控制compression前后精度在一定范围。
We propose a novel Convolutional Neural Network (CNN) compression algorithm based on coreset representations of filters. We exploit the redundancies extant in the space of CNN weights and neu- ronal activations (across samples) in order to obtain compression. Our method requires no retraining, is ...