English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Plus annual

Presents the access of premium content as premium feature

Premium Content

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Plus monthly

Global: Zendy Tools annual

Global: Zendy Tools monthly

Global: Zendy Free Plan

Inference for Deep Neural Networks is increasingly being executed locally onmobile and embedded platforms due to its advantages in latency, privacy andconnectivity. Since modern System on Chips typically execute a combination ofdifferent and dynamic workloads concurrently, it is challenging to consistentlymeet inference time/energy budget at runtime because of the local computingresources available to the DNNs vary considerably. To address this challenge, avariety of dynamic DNNs were proposed. However, these works have significantmemory overhead, limited runtime recoverable compression rate and narrowdynamic ranges of performance scaling. In this paper, we present a dynamic DNNusing incremental training and group convolution pruning. The channels of theDNN convolution layer are divided into groups, which are then trainedincrementally. At runtime, following groups can be pruned for inferencetime/energy reduction or added back for accuracy recovery without modelretraining. In addition, we combine task mapping and Dynamic Voltage FrequencyScaling (DVFS) with our dynamic DNN to deliver finer trade-off between accuracyand time/power/energy over a wider dynamic range. We illustrate the approach bymodifying AlexNet for the CIFAR10 image dataset and evaluate our work on twoheterogeneous hardware platforms: Odroid XU3 (ARM big.LITTLE CPUs) and NvidiaJetson Nano (CPU and GPU). Compared to the existing works, our approach canprovide up to 2.36x (energy) and 2.73x (time) wider dynamic range with a 2.4xsmaller memory footprint at the same compression rate. It achieved 10.6x(energy) and 41.6x (time) wider dynamic range by combining with task mappingand DVFS.

Incremental Training and Group Convolution Pruning for Runtime DNN Performance Scaling on Heterogeneous Embedded Platforms