Sparsified SGD with Memory

2018/09/20 by Sebastian U. Stich, Stich, Sebastian U., Jean-Baptiste Cordonnier +3 · 48 citations
Computer Science · Engineering · #68W15 #68W40 #90C06 #90C25 #Data Structures and Algorithms (cs.DS) #Distributed #Distributed Sensor Networks and Detection Algorithms #E.4 #F.2.1 #FOS: Computer and information sciences #G.1.6 #Machine Learning (cs.LG) #Machine Learning (stat.ML) #Parallel #Sparse and Compressive Sensing Techniques #Stochastic Gradient Optimization Techniques #and Cluster Computing (cs.DC)

paper · pdf · doi:10.48550/arxiv.1809.07599

openalex publication_date 2018/09/20 · openalex created_date 2025/10/10 · openalex updated_date 2026/07/28

Abstract

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 sparsification techniques to reduce the amount of data that needs to be communicated, for instance by only sending the most significant entries of the stochastic gradient (top-k sparsification). Whilst such schemes showed very promising performance in practice, they have eluded theoretical analysis so far. In this work we analyze Stochastic Gradient Descent (SGD) with k-sparsification or compression (for instance top-k or random-k) and show that this scheme converges at the same rate as vanilla SGD when equipped with error compensation (keeping track of accumulated errors in memory). That is, communication can be reduced by a factor of the dimension of the problem (sometimes even more) whilst still converging at the same rate. We present numerical experiments to illustrate the theoretical findings and the better scalability for distributed applications.

Cited by

• OptiNIC: A Resilient and Tail-Optimal RDMA NIC for Distributed ML Workloads
• Communication Compression for Distributed Learning with Aggregate and Server-Guided Feedback
• Timely Parameter Updating in Over-the-Air Federated Learning
• ACE-Sync: An Adaptive Cloud-Edge Synchronization Framework for Communication-Efficient Large-Scale Distributed Model Training
• A Systems-Theoretic View on the Convergence of Algorithms under Disturbances
• DRIVE: One-bit Distributed Mean Estimation
• DP-CSGP: Differentially Private Stochastic Gradient Push with Compressed Communication
• CurvaDion: Curvature-Adaptive Distributed Orthonormalization
• Distributed Online Randomized Gradient-Free Optimization with Compressed Communication
• Efficient Covariance Estimation for Sparsified Functional Data
• On the Gradient Complexity of Private Optimization with Private Oracles
• An Efficient Gradient-Aware Error-Bounded Lossy Compressor for Federated Learning
• TT-Prune: Joint Model Pruning and Resource Allocation for Communication-efficient Time-triggered Federated Learning
• Gradient Projection onto Historical Descent Directions for Communication-Efficient Federated Learning
• An All-Reduce Compatible Top-K Compressor for Communication-Efficient Distributed Learning
• Lightweight Federated Learning in Mobile Edge Computing with Statistical and Device Heterogeneity Awareness
• SpeechLLM Meets Federated Learning for End-to-End ASR: English and Italian Case Studies
• Unbiased Gradient Low-Rank Projection
• SketchGuard: Scaling Byzantine-Robust Decentralized Federated Learning via Sketch-Based Screening
• Composite Optimization with Error Feedback: the Dual Averaging Approach
• FedMuon: Federated Learning with Bias-corrected LMO-based Optimization
• Reconciling Communication Compression and Byzantine-Robustness in Distributed Learning
• From PowerSGD to PowerSGD+: Low-Rank Gradient Compression for Distributed Optimization with Convergence Guarantees
• Experiments with Rich Regime Training for Deep Learning
• Delayed Momentum Aggregation: Communication-efficient Byzantine-robust Federated Learning with Partial Participation
• DPQuant: Efficient and Differentially-Private Model Training via Dynamic Quantization Scheduling
• Elastic Consistency: A General Consistency Model for Distributed Stochastic Gradient Descent
• Masked Training of Neural Networks with Partial Gradients
• Distributed Online Learning for Joint Regret with Communication Constraints
• Quantized Frank-Wolfe: Faster Optimization, Lower Communication, and Projection Free
• To Talk or to Work: Flexible Communication Compression for Energy Efficient Federated Learning over Heterogeneous Mobile Edge Devices
• Solon: Communication-efficient Byzantine-resilient Distributed Training via Redundant Gradients
• Norm-Constrained Flows and Sign-Based Optimization: Theory and Algorithms
• Permutation Compressors for Provably Faster Distributed Nonconvex Optimization
• A Field Guide to Federated Optimization
• Communication-Efficient Distributed Asynchronous ADMM
• MergeComp: A Compression Scheduler for Scalable Communication-Efficient Distributed Training
• Resource-Aware Aggregation and Sparsification in Heterogeneous Ensemble Federated Learning
• Decentralized Relaxed Smooth Optimization with Gradient Descent Methods
• Decentralized Composite Optimization with Compression
• HeteRo-Select: Informativeness as the Participation Driver in Heterogeneous Federated Learning
• Distributed Optimization and Learning for Automated Stepsize Selection with Finite Time Coordination
• Compressed Decentralized Momentum Stochastic Gradient Methods for Nonconvex Optimization
• Federated Learning with Feature Reconstruction for Vector Quantization based Semantic Communication
• Feature Reconstruction Aided Federated Learning for Image Semantic Communication
• Energy-Efficient Federated Learning for Edge Real-Time Vision via Joint Data, Computation, and Communication Design
• Communication-efficient distributed SGD with Sketching
• A Better Alternative to Error Feedback for Communication-Efficient\n Distributed Learning

Related

• Understanding Top-k Sparsification in Distributed Deep Learning
• Sparse Communication for Training Deep Networks
• Distributed Sparse SGD with Majority Voting
• Unbiased Single-scale and Multi-scale Quantizers for Distributed Optimization
• Gradient Sparification for Asynchronous Distributed Training