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

Driven by significant improvements in architectural design and trainingpipelines, computer vision has recently experienced dramatic progress in termsof accuracy on classic benchmarks such as ImageNet. These highly-accuratemodels are challenging to deploy, as they appear harder to compress usingstandard techniques such as pruning. We address this issue by introducing theCorrelation Aware Pruner (CAP), a new unstructured pruning framework whichsignificantly pushes the compressibility limits for state-of-the-artarchitectures. Our method is based on two technical advancements: a newtheoretically-justified pruner, which can handle complex weight correlationsaccurately and efficiently during the pruning process itself, and an efficientfinetuning procedure for post-compression recovery. We validate our approachvia extensive experiments on several modern vision models such as VisionTransformers (ViT), modern CNNs, and ViT-CNN hybrids, showing for the firsttime that these can be pruned to high sparsity levels (e.g. $\geq 75$%) withlow impact on accuracy ($\leq 1$% relative drop). Our approach is alsocompatible with structured pruning and quantization, and can lead to practicalspeedups of 1.5 to 2.4x without accuracy loss. To further showcase CAP'saccuracy and scalability, we use it to show for the first time thatextremely-accurate large vision models, trained via self-supervised techniques,can also be pruned to moderate sparsities, with negligible accuracy loss.

CAP: Correlation-Aware Pruning for Highly-Accurate Sparse Vision Models