arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2412.18052cs.LGcs.AI
收藏

超越并行优化中的梯度平均:通过梯度一致性过滤提升鲁棒性

Beyond Gradient Averaging in Parallel Optimization: Improved Robustness through Gradient Agreement Filtering

Stanford University(斯坦福大学)

Francois Chaubard, Duncan Eddy, Mykel J. Kochenderfer

AI总结:

本文提出梯度一致性过滤(GAF),通过在梯度平均前计算微梯度余弦距离并过滤冲突更新,降低梯度方差,从而以更小微批次提升验证准确率、减少噪声标签记忆,并在CIFAR基准上实现最高18.2%的准确率提升和近一个数量级的计算量下降。

摘要:

我们提出梯度一致性过滤(Gradient Agreement Filtering,GAF),以改进分布式深度学习优化中的梯度平均方法。传统的分布式数据并行随机梯度下降通过对微批次(microbatches)的梯度进行平均来计算宏批次(macrobatch)梯度,然后用该梯度更新模型参数。我们发现,不同微批次之间的梯度常常正交或负相关,尤其是在训练后期,这会导致对训练集的记忆,从而降低泛化能力。本文提出一种简单且计算高效的方法,通过在训练过程中计算微梯度之间的余弦距离,并在平均之前过滤掉相互冲突的更新,从而降低梯度方差。我们以显著更小的微批次规模提高了验证准确率。我们还表明,该方法能够减少对噪声标签的记忆。我们在标准图像分类基准(包括 CIFAR-100 和 CIFAR-100N-Fine)上验证了该技术的有效性。结果表明,与传统训练方法相比,该技术持续提升验证准确率,在某些情况下提升幅度高达 18.2%,同时将所需计算量减少近一个数量级,因为我们现在可以依赖更小的微批次规模而不会破坏训练稳定性。

英文摘要:

We introduce Gradient Agreement Filtering (GAF) to improve on gradient averaging in distributed deep learning optimization. Traditional distributed data-parallel stochastic gradient descent involves averaging gradients of microbatches to calculate a macrobatch gradient that is then used to update model parameters. We find that gradients across microbatches are often orthogonal or negatively correlated, especially in late stages of training, which leads to memorization of the training set, reducing generalization. In this paper, we introduce a simple, computationally effective way to reduce gradient variance by computing the cosine distance between micro-gradients during training and filtering out conflicting updates prior to averaging. We improve validation accuracy with significantly smaller microbatch sizes. We also show this reduces memorizing noisy labels. We demonstrate the effectiveness of this technique on standard image classification benchmarks including CIFAR-100 and CIFAR-100N-Fine. We show this technique consistently outperforms validation accuracy, in some cases by up to 18.2\% compared to traditional training approaches while reducing the computation required nearly an order of magnitude because we can now rely on smaller microbatch sizes without destabilizing training.

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。

↑