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 1907.04840cs.LGcs.NEstat.ML
收藏

从零开始稀疏网络:更快的训练而不损失性能

Sparse Networks from Scratch: Faster Training without Losing Performance

Tim Dettmers, Luke Zettlemoyer

摘要:

我们展示了我们称之为稀疏学习的可能性:深度神经网络的加速训练,在整个训练过程中保持稀疏权重,同时达到密集的性能水平。我们通过开发稀疏动量来实现这一点,稀疏动量是一种使用指数平滑梯度(动量)来识别有效减少错误的层和权重的算法。稀疏动量根据每层的平均动量大小在层间重新分配修剪后的权重。在一个层中,稀疏动量根据零值权重的动量大小来增加权重。我们在MNIST、CIFAR-10和ImageNet上展示了最先进的稀疏性能,与其他稀疏算法相比,平均误差分别降低了8%、15%和6%。此外,我们还证明了稀疏动量可以可靠地再现密集的性能水平,同时提供高达5.61倍的训练速度。在我们的分析中,消融表明,动量重新分配和增长的好处随着网络的深度和规模的增加而增加。此外,我们发现稀疏动量对超参数的选择不敏感,这表明稀疏动量是鲁棒的,易于使用。

英文摘要:

We demonstrate the possibility of what we call sparse learning: accelerated training of deep neural networks that maintain sparse weights throughout training while achieving dense performance levels. We accomplish this by developing sparse momentum, an algorithm which uses exponentially smoothed gradients (momentum) to identify layers and weights which reduce the error efficiently. Sparse momentum redistributes pruned weights across layers according to the mean momentum magnitude of each layer. Within a layer, sparse momentum grows weights according to the momentum magnitude of zero-valued weights. We demonstrate state-of-the-art sparse performance on MNIST, CIFAR-10, and ImageNet, decreasing the mean error by a relative 8%, 15%, and 6% compared to other sparse algorithms. Furthermore, we show that sparse momentum reliably reproduces dense performance levels while providing up to 5.61x faster training. In our analysis, ablations show that the benefits of momentum redistribution and growth increase with the depth and size of the network. Additionally, we find that sparse momentum is insensitive to the choice of its hyperparameters suggesting that sparse momentum is robust and easy to use.

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。

↑