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 Tools annual

Global: Zendy Free Plan

Global: Zendy Tools monthly

Global: Zendy Plus monthly

With the emergence of a spectrum of high-end mobile devices, manyapplications that formerly required desktop-level computation capability arebeing transferred to these devices. However, executing the inference of DeepNeural Networks (DNNs) is still challenging considering high computation andstorage demands, specifically, if real-time performance with high accuracy isneeded. Weight pruning of DNNs is proposed, but existing schemes represent twoextremes in the design space: non-structured pruning is fine-grained, accurate,but not hardware friendly; structured pruning is coarse-grained,hardware-efficient, but with higher accuracy loss. In this paper, we introducea new dimension, fine-grained pruning patterns inside the coarse-grainedstructures, revealing a previously unknown point in design space. With thehigher accuracy enabled by fine-grained pruning patterns, the unique insight isto use the compiler to re-gain and guarantee high hardware efficiency. In otherwords, our method achieves the best of both worlds, and is desirable acrosstheory/algorithm, compiler, and hardware levels. The proposed PatDNN is anend-to-end framework to efficiently execute DNN on mobile devices with the helpof a novel model compression technique (pattern-based pruning based on extendedADMM solution framework) and a set of thorough architecture-aware compiler- andcode generation-based optimizations (filter kernel reordering, compressedweight storage, register load redundancy elimination, and parameterauto-tuning). Evaluation results demonstrate that PatDNN outperforms threestate-of-the-art end-to-end DNN frameworks, TensorFlow Lite, TVM, and AlibabaMobile Neural Network with speedup up to 44.5x, 11.4x, and 7.1x, respectively,with no accuracy compromise. Real-time inference of representative large-scaleDNNs (e.g., VGG-16, ResNet-50) can be achieved using mobile devices.

PatDNN: Achieving Real-Time DNN Execution on Mobile Devices with Pattern-based Weight Pruning