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

Most of today's computer vision pipelines are built around deep neuralnetworks, where convolution operations require most of the generally highcompute effort. The Winograd convolution algorithm computes convolutions withfewer MACs compared to the standard algorithm, reducing the operation count bya factor of 2.25x for 3x3 convolutions when using the version with 2x2-sizedtiles $F_2$. Even though the gain is significant, the Winograd algorithm withlarger tile sizes, i.e., $F_4$, offers even more potential in improvingthroughput and energy efficiency, as it reduces the required MACs by 4x.Unfortunately, the Winograd algorithm with larger tile sizes introducesnumerical issues that prevent its use on integer domain-specific acceleratorsand higher computational overhead to transform input and output data betweenspatial and Winograd domains.  To unlock the full potential of Winograd $F_4$, we propose a novel tap-wisequantization method that overcomes the numerical issues of using larger tiles,enabling integer-only inference. Moreover, we present custom hardware unitsthat process the Winograd transformations in a power- and area-efficient way,and we show how to integrate such custom modules in an industrial-grade,programmable DSA. An extensive experimental evaluation on a large set ofstate-of-the-art computer vision benchmarks reveals that the tap-wisequantization algorithm makes the quantized Winograd $F_4$ network almost asaccurate as the FP32 baseline. The Winograd-enhanced DSA achieves up to 1.85xgain in energy efficiency and up to 1.83x end-to-end speed-up forstate-of-the-art segmentation and detection networks.

Going Further With Winograd Convolutions: Tap-Wise Quantization for Efficient Inference on 4x4 Tile