Adaptive extreme edge computing for wearable devices
Wearable devices are a fast-growing technology with impact on personal healthcare for both
society and economy. Due to the widespread of sensors in pervasive and distributed …
society and economy. Due to the widespread of sensors in pervasive and distributed …
Communication-efficient distributed deep learning: A comprehensive survey
Distributed deep learning (DL) has become prevalent in recent years to reduce training time
by leveraging multiple computing devices (eg, GPUs/TPUs) due to larger models and …
by leveraging multiple computing devices (eg, GPUs/TPUs) due to larger models and …
Sparsity in deep learning: Pruning and growth for efficient inference and training in neural networks
The growing energy and performance costs of deep learning have driven the community to
reduce the size of neural networks by selectively pruning components. Similarly to their …
reduce the size of neural networks by selectively pruning components. Similarly to their …
Rigging the lottery: Making all tickets winners
Many applications require sparse neural networks due to space or inference time
restrictions. There is a large body of work on training dense networks to yield sparse …
restrictions. There is a large body of work on training dense networks to yield sparse …
The state of sparsity in deep neural networks
We rigorously evaluate three state-of-the-art techniques for inducing sparsity in deep neural
networks on two large-scale learning tasks: Transformer trained on WMT 2014 English-to …
networks on two large-scale learning tasks: Transformer trained on WMT 2014 English-to …
Accelerating sparse deep neural networks
As neural network model sizes have dramatically increased, so has the interest in various
techniques to reduce their parameter counts and accelerate their execution. An active area …
techniques to reduce their parameter counts and accelerate their execution. An active area …
Train big, then compress: Rethinking model size for efficient training and inference of transformers
Since hardware resources are limited, the objective of training deep learning models is
typically to maximize accuracy subject to the time and memory constraints of training and …
typically to maximize accuracy subject to the time and memory constraints of training and …
[HTML][HTML] Scalable distributed DNN training using commodity GPU cloud computing
N Ström - 2015 - amazon.science
We introduce a new method for scaling up distributed Stochastic Gradient Descent (SGD)
training of Deep Neural Networks (DNN). The method solves the well-known communication …
training of Deep Neural Networks (DNN). The method solves the well-known communication …
Top-kast: Top-k always sparse training
Sparse neural networks are becoming increasingly important as the field seeks to improve
the performance of existing models by scaling them up, while simultaneously trying to …
the performance of existing models by scaling them up, while simultaneously trying to …
Powerpropagation: A sparsity inducing weight reparameterisation
The training of sparse neural networks is becoming an increasingly important tool for
reducing the computational footprint of models at training and evaluation, as well enabling …
reducing the computational footprint of models at training and evaluation, as well enabling …