arXivDaily arXiv每日学术速递 周一至周五更新
arXiv周末暂无论文更新,休息一下吧,周末愉快~~
arXiv 2007.00389cs.LGstat.ML
收藏

训练前的Single Shot结构化剪枝

Single Shot Structured Pruning Before Training

University of Oxford(牛津大学)

Joost van Amersfoort, Milad Alizadeh, Sebastian Farquhar, Nicholas Lane, Yarin Gal

摘要:

我们介绍了一种使用在训练前应用的结构化修剪来将深度神经网络中的训练速度加快2倍并将推理速度加快3倍的方法。与以前在训练前修剪单个权重的工作不同,我们的工作开发了一种方法来删除整个通道和隐藏单元,其明确目标是加快训练和推理。我们引入了一个计算感知的评分机制,使修剪单位的灵敏度每FLOP删除,允许更大的速度。我们的方法快速,易于实现,并且在训练开始之前只需要对一批数据进行一次向前/向后传递即可完成修剪。

英文摘要:

We introduce a method to speed up training by 2x and inference by 3x in deep neural networks using structured pruning applied before training. Unlike previous works on pruning before training which prune individual weights, our work develops a methodology to remove entire channels and hidden units with the explicit aim of speeding up training and inference. We introduce a compute-aware scoring mechanism which enables pruning in units of sensitivity per FLOP removed, allowing even greater speed ups. Our method is fast, easy to implement, and needs just one forward/backward pass on a single batch of data to complete pruning before training begins.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑