论文标题
一个DNN中有几张获胜的门票？
How many winning tickets are there in one DNN?
论文作者
论文摘要
最近的彩票假设提出，有一个子网与隔离训练时与原始网络的准确性相匹配。我们表明，即使固定了初始权重，每个网络都包含几个获胜门票。由此产生的获胜子网络并不是体重空间对称性下同一网络的实例，并且没有明显的重叠或相关性明显大于偶然的预期。如果训练期间的随机性减少，即使网络受到不同任务的培训，重叠也高于机会。我们得出的结论是，与单个获胜的票相反，有能力的子网络有一个分布。
The recent lottery ticket hypothesis proposes that there is one sub-network that matches the accuracy of the original network when trained in isolation. We show that instead each network contains several winning tickets, even if the initial weights are fixed. The resulting winning sub-networks are not instances of the same network under weight space symmetry, and show no overlap or correlation significantly larger than expected by chance. If randomness during training is decreased, overlaps higher than chance occur, even if the networks are trained on different tasks. We conclude that there is rather a distribution over capable sub-networks, as opposed to a single winning ticket.
