BriefGPT.xyz
May, 2018
深度神经网络的推理阶段统一和合并
Unifying and Merging Well-trained Deep Neural Networks for Inference Stage
HTML
PDF
Yi-Min Chou, Yi-Ming Chan, Jia-Hong Lee, Chih-Yi Chiu, Chu-Song Chen
TL;DR
本文提出了一种新的卷积神经网络合并方法，在共享权重的基础上，将训练好的两个具有不同架构的网络在推断阶段融合为一个统一的模型，以适应资源有限的设备，实验结果表明此方法有效。
Abstract
We propose a novel method to merge
convolutional neural-nets
for the
inference
stage. Given two well-trained networks that may have different
arc
→