DeepAI AI Chat
Log In Sign Up

Adaptive Braking for Mitigating Gradient Delay

by   Abhinav Venigalla, et al.

Neural network training is commonly accelerated by using multiple synchronized workers to compute gradient updates in parallel. Asynchronous methods remove synchronization overheads and improve hardware utilization at the cost of introducing gradient delay, which impedes optimization and can lead to lower final model performance. We introduce Adaptive Braking (AB), a modification for momentum-based optimizers that mitigates the effects of gradient delay. AB dynamically scales the gradient based on the alignment of the gradient and the velocity. This can dampen oscillations along high curvature directions of the loss surface, stabilizing and accelerating asynchronous training. We show that applying AB on top of SGD with momentum enables training ResNets on CIFAR-10 and ImageNet-1k with delays D ≥ 32 update steps with minimal drop in final test accuracy.


page 2

page 13

page 14


Decaying momentum helps neural network training

Momentum is a simple and popular technique in deep learning for gradient...

A Mechanism for Distributed Deep Learning Communication Optimization

Intensive communication and synchronization cost for gradients and param...

Taming Momentum in a Distributed Asynchronous Environment

Although distributed computing can significantly reduce the training tim...

Gap Aware Mitigation of Gradient Staleness

Cloud computing is becoming increasingly popular as a platform for distr...

Aggregated Momentum: Stability Through Passive Damping

Momentum is a simple and widely used trick which allows gradient-based o...

Delay-adaptive step-sizes for asynchronous learning

In scalable machine learning systems, model training is often paralleliz...

Accumulated Decoupled Learning: Mitigating Gradient Staleness in Inter-Layer Model Parallelization

Decoupled learning is a branch of model parallelism which parallelizes t...