arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 1712.04432cs.LGstat.ML
收藏

神经网络训练中的模型、批处理和领域并行性集成

Integrated Model, Batch and Domain Parallelism in Training Neural Networks

UC Berkeley(加州大学伯克利分校);Lawrence Berkeley Lab(劳伦斯伯克利实验室)

Amir Gholami, Ariful Azad, Peter Jin, Kurt Keutzer, Aydin Buluc

摘要:

我们提出了一种新的集成方法,利用模型,批处理和域并行性,使用小批量随机梯度下降(SGD)在大型分布式内存计算机上训练深度神经网络(DNN)。我们的目标是找到一个有效的并行化策略,一个固定的批量大小使用$P$进程。我们的方法的灵感来自于数值线性代数中的通信避免算法。我们将$P$流程在逻辑上划分为一个$P_r \times P_c$网格,其中$P_r$维隐式地负责模型/域并行性,而$P_c$维隐式地负责批处理并行性。在实践中,集成的基于矩阵的并行算法自动封装这些类型的并行性。我们分析了通信复杂性,并分析表明,最低的通信成本往往既不与纯模型,也不与纯数据并行。我们还展示了如何域并行方法可以帮助在扩展的理论规模限制的典型的批处理并行方法。

英文摘要:

We propose a new integrated method of exploiting model, batch and domain parallelism for the training of deep neural networks (DNNs) on large distributed-memory computers using minibatch stochastic gradient descent (SGD). Our goal is to find an efficient parallelization strategy for a fixed batch size using $P$ processes. Our method is inspired by the communication-avoiding algorithms in numerical linear algebra. We see $P$ processes as logically divided into a $P_r \times P_c$ grid where the $P_r$ dimension is implicitly responsible for model/domain parallelism and the $P_c$ dimension is implicitly responsible for batch parallelism. In practice, the integrated matrix-based parallel algorithm encapsulates these types of parallelism automatically. We analyze the communication complexity and analytically demonstrate that the lowest communication costs are often achieved neither with pure model nor with pure data parallelism. We also show how the domain parallel approach can help in extending the theoretical scaling limit of the typical batch parallel method.

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。

↑