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 Tools annual

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 monthly

Global: Zendy Plus monthly

Presents the access of premium content as premium feature

Premium Content

Global: Zendy Plus annual

Global: Zendy Free Plan

Deep neural networks have achieved impressive results in computer vision andmachine learning. Unfortunately, state-of-the-art networks are extremelycompute and memory intensive which makes them unsuitable for mW-devices such asIoT end-nodes. Aggressive quantization of these networks dramatically reducesthe computation and memory footprint. Binary-weight neural networks (BWNs)follow this trend, pushing weight quantization to the limit. Hardwareaccelerators for BWNs presented up to now have focused on core efficiency,disregarding I/O bandwidth and system-level efficiency that are crucial fordeployment of accelerators in ultra-low power devices. We present Hyperdrive: aBWN accelerator dramatically reducing the I/O bandwidth exploiting a novelbinary-weight streaming approach, which can be used for arbitrarily sizedconvolutional neural network architecture and input resolution by exploitingthe natural scalability of the compute units both at chip-level andsystem-level by arranging Hyperdrive chips systolically in a 2D mesh whileprocessing the entire feature map together in parallel. Hyperdrive achieves 4.3TOp/s/W system-level efficiency (i.e., including I/Os)---3.1x higher thanstate-of-the-art BWN accelerators, even if its core uses resource-intensiveFP16 arithmetic for increased robustness.

Hyperdrive: A Multi-Chip Systolically Scalable Binary-Weight CNN Inference Engine