Faster Neural Network Training with Data Echoing

Abstract

In the twilight of Moore's law, GPUs and other specialized hardware accelerators have dramatically sped up neural network training. However, earlier stages of the training pipeline, such as disk I/O and data preprocessing, do not run on accelerators. As accelerators continue to improve, these earlier stages will increasingly become the bottleneck. In this paper, we introduce "data echoing," which reduces the total computation used by earlier pipeline stages and speeds up training whenever computation upstream from accelerators dominates the training time. Data echoing reuses (or "echoes") intermediate outputs from earlier pipeline stages in order to reclaim idle capacity. We investigate the behavior of different data echoing algorithms on various workloads, for various amounts of echoing, and for various batch sizes. We find that in all settings, at least one data echoing algorithm can match the baseline's predictive performance using less upstream computation. We measured a factor of 3.25 decrease in wall-clock time for ResNet-50 on ImageNet when reading training data over a network.

References in corpus (9)

• Batch Normalization: Accelerating Deep Network Training by Reducing Internal Covariate Shift
• Measuring the Effects of Data Parallelism on Neural Network Training
• Not All Samples Are Created Equal: Deep Learning with Importance Sampling
• Image Classification at Supercomputer Scale
• Parallel SGD: When does averaging help?
• Characterizing Deep-Learning I/O Workloads in TensorFlow
• Augment your batch: better training with larger batches
• Critical Hyper-Parameters: No Random, No Cry
• Faster SGD training by minibatch persistency

Cited by in corpus (11)

• Understanding Data Storage and Ingestion for Large-Scale Deep Recommendation Model Training
• Alleviating the transit timing variation bias in transit surveys. I. RIVERS: Method and detection of a pair of resonant super-Earths around Kepler-1705
• tf-Darshan: Understanding Fine-grained I/O Performance in Machine Learning Workloads
• tf.data: A Machine Learning Data Processing Framework
• Plumber: Diagnosing and Removing Performance Bottlenecks in Machine Learning Data Pipelines
• Data augmentation in Bayesian neural networks and the cold posterior effect
• Drawing Multiple Augmentation Samples Per Image During Training Efficiently Decreases Test Error
• Stochastic Optimization with Laggard Data Pipelines
• Parameter Prediction for Unseen Deep Architectures
• Acceleration via Fractal Learning Rate Schedules
• Progressive Compressed Records: Taking a Byte out of Deep Learning Data