Accelerating Deep Learning by Focusing on the Biggest Losers

Abstract

This paper introduces Selective-Backprop, a technique that accelerates the training of deep neural networks (DNNs) by prioritizing examples with high loss at each iteration. Selective-Backprop uses the output of a training example's forward pass to decide whether to use that example to compute gradients and update parameters, or to skip immediately to the next example. By reducing the number of computationally-expensive backpropagation steps performed, Selective-Backprop accelerates training. Evaluation on CIFAR10, CIFAR100, and SVHN, across a variety of modern image models, shows that Selective-Backprop converges to target error rates up to 3.5x faster than with standard SGD and between 1.02--1.8x faster than a state-of-the-art importance sampling approach. Further acceleration of 26% can be achieved by using stale forward pass results for selection, thus also skipping forward passes of low priority examples.

References in corpus (8)

• Improved Regularization of Convolutional Neural Networks with Cutout
• Wide Residual Networks
• Deep k-Nearest Neighbors: Towards Confident, Interpretable and Robust Deep Learning
• Online Batch Selection for Faster Training of Neural Networks
• Variance Reduction in SGD by Distributed Importance Sampling
• Biased Importance Sampling for Deep Neural Network Training
• Active Sampler: Light-weight Accelerator for Complex Data Analytics at Scale
• AutoAssist: A Framework to Accelerate Training of Deep Neural Networks

Cited by in corpus (13)

• Towards the Systematic Reporting of the Energy and Carbon Footprints of Machine Learning
• Client Selection in Federated Learning: Convergence Analysis and Power-of-Choice Selection Strategies
• A Study of Gradient Variance in Deep Learning
• On Tilted Losses in Machine Learning: Theory and Applications
• L${}^2$-GCN: Layer-Wise and Learned Efficient Training of Graph Convolutional Networks
• When Do Curricula Work?
• Tilted Empirical Risk Minimization
• Enabling On-Device Self-Supervised Contrastive Learning With Selective Data Contrast
• When does loss-based prioritization fail?
• One Backward from Ten Forward, Subsampling for Large-Scale Deep Learning
• Trivial or impossible -- dichotomous data difficulty masks model differences (on ImageNet and beyond)
• Training With Data Dependent Dynamic Learning Rates
• Training Efficiency and Robustness in Deep Learning