arXiv research

Adaptive step-size method improves compressed SGD performance in machine learning.

Sign-based algorithms (e.g. signSGD) have been proposed as a biased gradient compression technique to alleviate the communication bottleneck in training large neural networks across multiple workers. We show simple convex counter-examples where signSGD does not converge to the optimum. Further, even when it does conver…

Paper improves compressed SGD to reach second-order stationary points.

IntSGD compresses SGD gradients without floats, converging as SGD.

SPARQ-SGD optimizes communication in decentralized SGD with event-triggered and compressed updates.

New SGD variant makes neural networks compressible without assumptions.

New study reveals how heavy-tailed SGD dynamics lead to compressible neural networks.

Unified analysis of SGD variants for nonconvex federated optimization.

Paper proposes LC-Checkpoint for efficient deep learning model checkpoints.

Communication overhead is a major bottleneck hampering the scalability of distributed machine learning systems. Recently, there has been a surge of interest in using gradient compression to improve the communication efficiency of distributed neural network training. Using 1-bit quantization, signSGD with majority vote …

We study gradient compression methods to alleviate the communication bottleneck in data-parallel distributed optimization. Despite the significant attention received, current compression schemes either do not scale well or fail to achieve the target test accuracy. We propose a new low-rank gradient compressor based on …

Communication bottleneck has been identified as a significant issue in distributed optimization of large-scale learning models. Recently, several approaches to mitigate this problem have been proposed, including different forms of gradient compression or computing local models and mixing them iteratively. In this paper…

We analyze (stochastic) gradient descent (SGD) with delayed updates on smooth quasi-convex and non-convex functions and derive concise, non-asymptotic, convergence rates. We show that the rate of convergence in all cases consists of two terms: (i) a stochastic term which is not affected by the delay, and (ii) a higher …

Adaptive quantization improves SGD accuracy in data-parallel settings.

New methods improve distributed optimization on non-iid data.

APMSqueeze improves Adam for faster training with less communication.

Analyze SGD with biased gradients, improving convergence rates and accuracy.

New method reduces communication in deep learning training.

SQuARM-SGD improves decentralized SGD efficiency with momentum.

CSER improves SGD efficiency by resetting errors and partial synchronization.

A new gradient quantization scheme improves communication efficiency in distributed training.

Proposes spred for solving $L_1$ penalty with SGD.

Due to its efficiency and ease to implement, stochastic gradient descent (SGD) has been widely used in machine learning. In particular, SGD is one of the most popular optimization methods for distributed learning. Recently, quantized SGD (QSGD), which adopts quantization to reduce the communication cost in SGD-based di…

We consider decentralized stochastic optimization with the objective function (e.g. data samples for machine learning task) being distributed over $n$ machines that can only communicate to their neighbors on a fixed communication graph. To reduce the communication bottleneck, the nodes compress (e.g. quantize or sparsi…

Unified framework for distributed compressed SGD under $(L_0, L_1)$-smoothness.

As the size and complexity of models and datasets grow, so does the need for communication-efficient variants of stochastic gradient descent that can be deployed to perform parallel model training. One popular communication-compression method for data-parallel SGD is QSGD (Alistarh et al., 2017), which quantizes and en…

Large-scale distributed training of neural networks is often limited by network bandwidth, wherein the communication time overwhelms the local computation time. Motivated by the success of sketching methods in sub-linear/streaming algorithms, we introduce Sketched SGD, an algorithm for carrying out distributed SGD by c…

Deep Neural Network (DNN) is powerful but computationally expensive and memory intensive, thus impeding its practical usage on resource-constrained front-end devices. DNN pruning is an approach for deep model compression, which aims at eliminating some parameters with tolerable performance degradation. In this paper, w…

Flexible framework compresses models using LC algorithm.

New algorithm resists Byzantine attacks in distributed SGD for heterogeneous data.

ResNet learns to compress information during training.

Large-scale distributed training requires significant communication bandwidth for gradient exchange that limits the scalability of multi-node training, and requires expensive high-bandwidth network infrastructure. The situation gets even worse with distributed training on mobile devices (federated learning), which suff…

A2SGD reduces distributed SGD communication to O(1) per worker.

Modern deep learning models are often trained in parallel over a collection of distributed machines to reduce training time. In such settings, communication of model updates among machines becomes a significant performance bottleneck and various lossy update compression techniques have been proposed to alleviate this p…

Stochastic gradient descent (SGD) is one of the most widely used optimization methods for parallel and distributed processing of large datasets. One of the key limitations of distributed SGD is the need to regularly communicate the gradients between different computation nodes. To reduce this communication bottleneck, …

Highly distributed training of Deep Neural Networks (DNNs) on future compute platforms (offering 100 of TeraOps/s of computational capacity) is expected to be severely communication constrained. To overcome this limitation, new gradient compression techniques are needed that are computationally friendly, applicable to …

Decentralized training of deep learning models is a key element for enabling data privacy and on-device learning over networks, as well as for efficient scaling to large compute clusters. As current approaches suffer from limited bandwidth of the network, we propose the use of communication compression in the decentral…

Proposes a method to improve Byzantine-robustness in compressed federated learning.

New method reduces communication costs in distributed nonconvex optimization.

SGD trains NNs to learn low-dimensional representations efficiently.

Two algorithms improve federated learning efficiency and resilience.

Data parallelism can boost the training speed of convolutional neural networks (CNN), but could suffer from significant communication costs caused by gradient aggregation. To alleviate this problem, several scalar quantization techniques have been developed to compress the gradients. But these techniques could perform …

New biased compression methods lead to faster convergence in distributed learning.

HSQ reduces communication costs in federated learning.

Unified sign-based compression for federated learning with faster convergence.

Huge scale machine learning problems are nowadays tackled by distributed optimization algorithms, i.e. algorithms that leverage the compute power of many devices for training. The communication overhead is a key bottleneck that hinders perfect scalability. Various recent works proposed to use quantization or sparsifica…

A new method reduces communication costs in distributed learning.

This paper analyzes error feedback in compressed federated learning for non-convex optimization problems.