Accordion: Adaptive Gradient Communication via Critical Learning Regime Identification

Abstract

Distributed model training suffers from communication bottlenecks due to frequent model updates transmitted across compute nodes. To alleviate these bottlenecks, practitioners use gradient compression techniques like sparsification, quantization, or low-rank updates. The techniques usually require choosing a static compression ratio, often requiring users to balance the trade-off between model accuracy and per-iteration speedup. In this work, we show that such performance degradation due to choosing a high compression ratio is not fundamental. An adaptive compression strategy can reduce communication while maintaining final test accuracy. Inspired by recent findings on critical learning regimes, in which small gradient errors can have irrecoverable impact on model performance, we propose Accordion a simple yet effective adaptive compression algorithm. While Accordion maintains a high enough compression rate on average, it avoids over-compressing gradients whenever in critical learning regimes, detected by a simple gradient-norm based criterion. Our extensive experimental study over a number of machine learning tasks in distributed environments indicates that Accordion, maintains similar model accuracy to uncompressed training, yet achieves up to 5.5x better compression and up to 4.1x end-to-end speedup over static approaches. We show that Accordion also works for adjusting the batch size, another popular strategy for alleviating communication bottlenecks.

References in corpus (18)

• Very Deep Convolutional Networks for Large-Scale Image Recognition
• PyTorch: An Imperative Style, High-Performance Deep Learning Library
• Accurate, Large Minibatch SGD: Training ImageNet in 1 Hour
• QSGD: Communication-Efficient SGD via Gradient Quantization and Encoding
• Horovod: fast and easy distributed deep learning in TensorFlow
• Measuring the Effects of Data Parallelism on Neural Network Training
• Gradient Descent Happens in a Tiny Subspace
• AdaBatch: Adaptive Batch Sizes for Training Deep Neural Networks
• Hessian-based Analysis of Large Batch Training and Robustness to Adversaries
• The Error-Feedback Framework: Better Rates for SGD with Delayed Gradients and Compressed Communication
• AdaComp : Adaptive Residual Gradient Compression for Data-Parallel Distributed Training
• Understanding Top-k Sparsification in Distributed Deep Learning
• On the Relation Between the Sharpest Directions of DNN Loss and the SGD Step Length
• The Early Phase of Neural Network Training
• On the Computational Inefficiency of Large Batch Sizes for Stochastic Gradient Descent
• Large batch size training of neural networks with adversarial training and second-order information
• The Break-Even Point on Optimization Trajectories of Deep Neural Networks
• Distributed Learning with Sublinear Communication

Cited by in corpus (9)

• Rethinking gradient sparsification as total error minimization
• AutoFreeze: Automatically Freezing Model Blocks to Accelerate Fine-tuning
• ScaDLES: Scalable Deep Learning over Streaming data at the Edge
• GraVAC: Adaptive Compression for Communication-Efficient Distributed DL Training
• Accelerating Distributed ML Training via Selective Synchronization
• Pufferfish: Communication-efficient Models At No Extra Cost
• Critical Learning Periods in Federated Learning
• Flexible Communication for Optimal Distributed Learning over Unpredictable Networks
• Solon: Communication-efficient Byzantine-resilient Distributed Training via Redundant Gradients