Breaking (Global) Barriers in Parallel Stochastic Optimization with Wait-Avoiding Group Averaging

• URL: http://arxiv.org/abs/2005.00124v3
• Date: Sat, 20 Feb 2021 15:36:09 GMT
• Title: Breaking (Global) Barriers in Parallel Stochastic Optimization with Wait-Avoiding Group Averaging
• Authors: Shigang Li, Tal Ben-Nun, Giorgi Nadiradze, Salvatore Di Girolamo, Nikoli Dryden, Dan Alistarh, Torsten Hoefler
• Abstract summary: We present WAGMA-SGD, a wait-avoiding subgroup that reduces global communication via weight exchange. We train ResNet-50 on ImageNet; Transformer for machine translation; and deep reinforcement learning for navigation at scale. Compared with state-of-the-art decentralized SGD variants, WAGMA-SGD significantly improves training throughput.
• Score: 34.55741812648229
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: Deep learning at scale is dominated by communication time. Distributing samples across nodes usually yields the best performance, but poses scaling challenges due to global information dissemination and load imbalance across uneven sample lengths. State-of-the-art decentralized optimizers mitigate the problem, but require more iterations to achieve the same accuracy as their globally-communicating counterparts. We present Wait-Avoiding Group Model Averaging (WAGMA) SGD, a wait-avoiding stochastic optimizer that reduces global communication via subgroup weight exchange. The key insight is a combination of algorithmic changes to the averaging scheme and the use of a group allreduce operation. We prove the convergence of WAGMA-SGD, and empirically show that it retains convergence rates similar to Allreduce-SGD. For evaluation, we train ResNet-50 on ImageNet; Transformer for machine translation; and deep reinforcement learning for navigation at scale. Compared with state-of-the-art decentralized SGD variants, WAGMA-SGD significantly improves training throughput (e.g., 2.1x on 1,024 GPUs for reinforcement learning), and achieves the fastest time-to-solution (e.g., the highest score using the shortest training time for Transformer).

Related papers

• Fast Graph Sharpness-Aware Minimization for Enhancing and Accelerating Few-Shot Node Classification [53.727688136434345]
  Graph Neural Networks (GNNs) have shown superior performance in node classification. We present Fast Graph Sharpness-Aware Minimization (FGSAM) that integrates the rapid training of Multi-Layer Perceptrons with the superior performance of GNNs. Our proposed algorithm outperforms the standard SAM with lower computational costs in FSNC tasks.
  arXiv  Detail & Related papers  (2024-10-22T09:33:29Z)
• ELGC-Net: Efficient Local-Global Context Aggregation for Remote Sensing Change Detection [65.59969454655996]
  We propose an efficient change detection framework, ELGC-Net, which leverages rich contextual information to precisely estimate change regions. Our proposed ELGC-Net sets a new state-of-the-art performance in remote sensing change detection benchmarks. We also introduce ELGC-Net-LW, a lighter variant with significantly reduced computational complexity, suitable for resource-constrained settings.
  arXiv  Detail & Related papers  (2024-03-26T17:46:25Z)
• Adaptive Sparse Convolutional Networks with Global Context Enhancement for Faster Object Detection on Drone Images [26.51970603200391]
  This paper investigates optimizing the detection head based on the sparse convolution. It suffers from inadequate integration of contextual information of tiny objects. We propose a novel global context-enhanced adaptive sparse convolutional network.
  arXiv  Detail & Related papers  (2023-03-25T14:42:50Z)
• Stochastic Weight Averaging Revisited [5.68481425260348]
  We show that SWA's performance is highly dependent on to what extent the SGD process that runs before SWA converges. We show that following an SGD process with insufficient convergence, running SWA more times leads to continual incremental benefits in terms of generalization.
  arXiv  Detail & Related papers  (2022-01-03T08:29:01Z)
• Gradient Coding with Dynamic Clustering for Straggler-Tolerant Distributed Learning [55.052517095437]
  gradient descent (GD) is widely employed to parallelize the learning task by distributing the dataset across multiple workers. A significant performance bottleneck for the per-iteration completion time in distributed synchronous GD is $straggling$ workers. Coded distributed techniques have been introduced recently to mitigate stragglers and to speed up GD iterations by assigning redundant computations to workers. We propose a novel dynamic GC scheme, which assigns redundant data to workers to acquire the flexibility to choose from among a set of possible codes depending on the past straggling behavior.
  arXiv  Detail & Related papers  (2021-03-01T18:51:29Z)
• Adaptive Periodic Averaging: A Practical Approach to Reducing Communication in Distributed Learning [6.370766463380455]
  We show that the optimal averaging period in terms of convergence and communication cost is not a constant, but instead varies over the course of the execution. We propose a practical algorithm, named ADaptive Periodic parameter averaging SGD (ADPSGD), to achieve a smaller overall variance of model parameters.
  arXiv  Detail & Related papers  (2020-07-13T00:04:55Z)
• AdamP: Slowing Down the Slowdown for Momentum Optimizers on Scale-invariant Weights [53.8489656709356]
  Normalization techniques are a boon for modern deep learning. It is often overlooked, however, that the additional introduction of momentum results in a rapid reduction in effective step sizes for scale-invariant weights. In this paper, we verify that the widely-adopted combination of the two ingredients lead to the premature decay of effective step sizes and sub-optimal model performances.
  arXiv  Detail & Related papers  (2020-06-15T08:35:15Z)
• O(1) Communication for Distributed SGD through Two-Level Gradient Averaging [0.0]
  We introduce a strategy called two-level gradient averaging (A2SGD) to consolidate all gradients down to merely two local averages per worker. Our theoretical analysis shows that A2SGD converges similarly like the default distributed SGD algorithm.
  arXiv  Detail & Related papers  (2020-06-12T18:20:52Z)
• DaSGD: Squeezing SGD Parallelization Performance in Distributed Training Using Delayed Averaging [4.652668321425679]
  Minibatch gradient descent (SGD) algorithm requires workers to halt forward/back propagations. DaSGD parallelizes SGD and forward/back propagations to hide 100% of the communication overhead.
  arXiv  Detail & Related papers  (2020-05-31T05:43:50Z)
• Communication-Efficient Distributed Stochastic AUC Maximization with Deep Neural Networks [50.42141893913188]
  We study a distributed variable for large-scale AUC for a neural network as with a deep neural network. Our model requires a much less number of communication rounds and still a number of communication rounds in theory. Our experiments on several datasets show the effectiveness of our theory and also confirm our theory.
  arXiv  Detail & Related papers  (2020-05-05T18:08:23Z)
• Detached Error Feedback for Distributed SGD with Random Sparsification [98.98236187442258]
  Communication bottleneck has been a critical problem in large-scale deep learning. We propose a new distributed error feedback (DEF) algorithm, which shows better convergence than error feedback for non-efficient distributed problems. We also propose DEFA to accelerate the generalization of DEF, which shows better bounds than DEF.
  arXiv  Detail & Related papers  (2020-04-11T03:50:59Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.