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

Albeit the scalable performance of vision transformers (ViTs), the densecomputational costs (training & inference) undermine their position inindustrial applications. Post-training quantization (PTQ), tuning ViTs with atiny dataset and running in a low-bit format, well addresses the cost issue butunluckily bears more performance drops in lower-bit cases. In this paper, weintroduce I&S-ViT, a novel method that regulates the PTQ of ViTs in aninclusive and stable fashion. I&S-ViT first identifies two issues in the PTQ ofViTs: (1) Quantization inefficiency in the prevalent log2 quantizer forpost-Softmax activations; (2) Rugged and magnified loss landscape incoarse-grained quantization granularity for post-LayerNorm activations. Then,I&S-ViT addresses these issues by introducing: (1) A novel shift-uniform-log2quantizer (SULQ) that incorporates a shift mechanism followed by uniformquantization to achieve both an inclusive domain representation and accuratedistribution approximation; (2) A three-stage smooth optimization strategy(SOS) that amalgamates the strengths of channel-wise and layer-wisequantization to enable stable learning. Comprehensive evaluations acrossdiverse vision tasks validate I&S-ViT' superiority over existing PTQ of ViTsmethods, particularly in low-bit scenarios. For instance, I&S-ViT elevates theperformance of 3-bit ViT-B by an impressive 50.68%.

I&S-ViT: An Inclusive & Stable Method for Pushing the Limit of Post-Training ViTs Quantization