Wall-clock convergence time and communication load are key performance metrics for the distributed implementation of stochastic gradient descent (SGD) in parameter server settings. Communication-adaptive distributed Adam (CADA) has been recently proposed as a way to reduce communication load via the adaptive selection of workers. CADA is subject to performance degradation in terms of wall-clock convergence time in the presence of stragglers. This paper proposes a novel scheme named grouping-based CADA (G-CADA) that retains the advantages of CADA in reducing the communication load, while increasing the robustness to stragglers at the cost of additional storage at the workers. G-CADA partitions the workers into groups of workers that are assigned the same data shards. Groups are scheduled adaptively at each iteration, and the server only waits for the fastest worker in each selected group. We provide analysis and experimental results to elaborate the significant gains on the wall-clock time, as well as communication load and computation load, of G-CADA over other benchmark schemes.
Distributed Stochastic Gradient Descent (SGD) when run in a synchronous manner, suffers from delays in waiting for the slowest learners (stragglers). Asynchronous methods can alleviate stragglers, but cause gradient staleness that can adversely affect convergence. In this work we present a novel theoretical characterization of the speed-up offered by asynchronous methods by analyzing the trade-off between the error in the trained model and the actual training runtime (wallclock time). The novelty in our work is that our runtime analysis considers random straggler delays, which helps us design …
SOTA decentralized SGD algorithms can overcome the bandwidth bottleneck at the parameter server by using communication collectives like Ring All-Reduce for synchronization. While the parameter updates in distributed SGD may happen asynchronously there is still a synchronization barrier to make sure that the local training epoch at every learner is complete before the learners can advance to the next epoch. The delays in waiting for the slowest learners(stragglers) remain to be a problem in the synchronization steps of these state-of-the-art decentralized frameworks. In this paper, we propose t…
Federated learning (FL) is a new machine learning framework which trains a joint model across a large amount of decentralized computing devices. Existing methods, e.g., Federated Averaging (FedAvg), are able to provide an optimization guarantee by synchronously training the joint model, but usually suffer from stragglers, i.e., IoT devices with low computing power or communication bandwidth, especially on heterogeneous optimization problems. To mitigate the influence of stragglers, this paper presents a novel FL algorithm, namely Hybrid Federated Learning (HFL), to achieve a learning balance i…
In this paper, we focus on approaches to parallelizing stochastic gradient descent (SGD) wherein data is farmed out to a set of workers, the results of which, after a number of updates, are then combined at a central master node. Although such synchronized SGD approaches parallelize well in idealized computing environments, they often fail to realize their promised computational acceleration in practical settings. One cause is slow workers, termed stragglers, who can cause the fusion step at the master node to stall, which greatly slowing convergence. In many straggler mitigation approaches wo…
Distributed Stochastic Gradient Descent (SGD) when run in a synchronous manner, suffers from delays in runtime as it waits for the slowest workers (stragglers). Asynchronous methods can alleviate stragglers, but cause gradient staleness that can adversely affect the convergence error. In this work, we present a novel theoretical characterization of the speedup offered by asynchronous methods by analyzing the trade-off between the error in the trained model and the actual training runtime(wallclock time). The main novelty in our work is that our runtime analysis considers random straggling dela…
Distributed stochastic gradient descent (SGD) is essential for scaling the machine learning algorithms to a large number of computing nodes. However, the infrastructures variability such as high communication delay or random node slowdown greatly impedes the performance of distributed SGD algorithm, especially in a wireless system or sensor networks. In this paper, we propose an algorithmic approach named Overlap-Local-SGD (and its momentum variant) to overlap the communication and computation so as to speedup the distributed training procedure. The approach can help to mitigate the straggler …
Distributed training of deep learning models on large-scale training data is typically conducted with asynchronous stochastic optimization to maximize the rate of updates, at the cost of additional noise introduced from asynchrony. In contrast, the synchronous approach is often thought to be impractical due to idle time wasted on waiting for straggling workers. We revisit these conventional beliefs in this paper, and examine the weaknesses of both approaches. We demonstrate that a third approach, synchronous optimization with backup workers, can avoid asynchronous noise while mitigating for th…
Distributed training of deep learning models on large-scale training data is typically conducted with asynchronous stochastic optimization to maximize the rate of updates, at the cost of additional noise introduced from asynchrony. In contrast, the synchronous approach is often thought to be impractical due to idle time wasted on waiting for straggling workers. We revisit these conventional beliefs in this paper, and examine the weaknesses of both approaches. We demonstrate that a third approach, synchronous optimization with backup workers, can avoid asynchronous noise while mitigating for th…
Communication cost is the main bottleneck for the design of effective distributed learning algorithms. Recently, event-triggered techniques have been proposed to reduce the exchanged information among compute nodes and thus alleviate the communication cost. However, most existing event-triggered approaches only consider heuristic event-triggered thresholds. They also ignore the impact of computation and network delay, which play an important role on the training performance. In this paper, we propose an Asynchronous Event-triggered Stochastic Gradient Descent (SGD) framework, called AET-SGD, t…
Technical University of Munich（慕尼黑工业大学）；Indian Institute of Science（印度科学学院）；US Army Research Laboratory（美国陆军研究实验室）；Rutgers University（罗格斯大学）
We consider the setting where a master wants to run a distributed stochastic gradient descent (SGD) algorithm on $n$ workers, each having a subset of the data. Distributed SGD may suffer from the effect of stragglers, i.e., slow or unresponsive workers who cause delays. One solution studied in the literature is to wait at each iteration for the responses of the fastest $k<n$ workers before updating the model, where $k$ is a fixed parameter. The choice of the value of $k$ presents a trade-off between the runtime (i.e., convergence rate) of SGD and the error of the model. Towards optimizing the …
Gradient-based distributed learning in Parameter Server (PS) computing architectures is subject to random delays due to straggling worker nodes, as well as to possible communication bottlenecks between PS and workers. Solutions have been recently proposed to separately address these impairments based on the ideas of gradient coding, worker grouping, and adaptive worker selection. This paper provides a unified analysis of these techniques in terms of wall-clock time, communication, and computation complexity measures. Furthermore, in order to combine the benefits of gradient coding and grouping…
Distributed training algorithms of deep neural networks show impressive convergence speedup properties on very large problems. However, they inherently suffer from communication related slowdowns and communication topology becomes a crucial design choice. Common approaches supported by most machine learning frameworks are: 1) Synchronous decentralized algorithms relying on a peer-to-peer All Reduce topology that is sensitive to stragglers and communication delays. 2) Asynchronous centralised algorithms with a server based topology that is prone to communication bottleneck. Researchers also sug…
LAYUP: Asynchronous decentralized gradient descent with LAYer-wise UPdates
Center for Cognitive Interaction Technology, Universität Bielefeld, Germany（认知交互技术中心，比勒菲尔德大学，德国）；Bielefeld University（比勒菲尔德大学）；Department of Computer Science, Royal Holloway, University of London, United Kingdom（计算机科学系，伦敦大学皇家霍洛威学院，英国）
We consider distributed gradient descent in the presence of stragglers. Recent work on \em gradient coding \em and \em approximate gradient coding \em have shown how to add redundancy in distributed gradient descent to guarantee convergence even if some workers are \em stragglers\em---that is, slow or non-responsive. In this work we propose an approximate gradient coding scheme called \em Stochastic Gradient Coding \em (SGC), which works when the stragglers are random. SGC distributes data points redundantly to workers according to a pair-wise balanced design, and then simply ignores the strag…
KAIST（韩国科学技术院）；IBM T.J. Watson Research Center（IBM托马斯·J·沃森研究中心）；AITRICS
Adaptive gradient approaches that automatically adjust the learning rate on a per-feature basis have been very popular for training deep networks. This rich class of algorithms includes Adagrad, RMSprop, Adam, and recent extensions. All these algorithms have adopted diagonal matrix adaptation, due to the prohibitive computational burden of manipulating full matrices in high-dimensions. In this paper, we show that block-diagonal matrix adaptation can be a practical and powerful solution that can effectively utilize structural characteristics of deep learning architectures, and significantly imp…
Facebook AI Research（Facebook人工智能研究院）；McGill University（麦吉尔大学）；University of Edinburgh（爱丁堡大学）
Distributed data-parallel algorithms aim to accelerate the training of deep neural networks by parallelizing the computation of large mini-batch gradient updates across multiple nodes. Approaches that synchronize nodes using exact distributed averaging (e.g., via AllReduce) are sensitive to stragglers and communication delays. The PushSum gossip algorithm is robust to these issues, but only performs approximate distributed averaging. This paper studies Stochastic Gradient Push (SGP), which combines PushSum with stochastic gradient updates. We prove that SGP converges to a stationary point of s…
Freya PAGE: First Optimal Time Complexity for Large-Scale Nonconvex Finite-Sum Optimization with Heterogeneous Asynchronous Computations
King Abdullah University of Science and Technology（阿卜杜拉国王科技大学）；AIRI；Skolkovo Institute of Science and Technology（斯科尔科沃科学技术研究院）；KAUST（阿卜杜拉国王科技大学）；Skoltech（斯科尔泰克科技大学）
Technion - Israel Institute of Technology（以色列理工学院）
Cloud computing is becoming increasingly popular as a platform for distributed training of deep neural networks. Synchronous stochastic gradient descent (SSGD) suffers from substantial slowdowns due to stragglers if the environment is non-dedicated, as is common in cloud computing. Asynchronous SGD (ASGD) methods are immune to these slowdowns but are scarcely used due to gradient staleness, which encumbers the convergence process. Recent techniques have had limited success mitigating the gradient staleness when scaling up to many workers (computing nodes). In this paper we define the Gap as a …
Worcester Polytechnic Institute（伍斯特理工学院）；State Key Lab of Computer Science, Institute of Software, Chinese Academy of Sciences（中国科学院软件研究所计算机科学国家重点实验室）
Stochastic Gradient Descent (SGD) has become the de facto way to train deep neural networks in distributed clusters. A critical factor in determining the training throughput and model accuracy is the choice of the parameter synchronization protocol. For example, while Bulk Synchronous Parallel (BSP) often achieves better converged accuracy, the corresponding training throughput can be negatively impacted by stragglers. In contrast, Asynchronous Parallel (ASP) can have higher throughput, but its convergence and accuracy can be impacted by stale gradients. To improve the performance of synchroni…
We study scheduling of computation tasks across n workers in a large scale distributed learning problem with the help of a master. Computation and communication delays are assumed to be random, and redundant computations are assigned to workers in order to tolerate stragglers. We consider sequential computation of tasks assigned to a worker, while the result of each computation is sent to the master right after its completion. Each computation round, which can model an iteration of the stochastic gradient descent (SGD) algorithm, is completed once the master receives k distinct computations, r…
