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

Quantization of deep neural networks (DNN) has been proven effective forcompressing and accelerating DNN models. Data-free quantization (DFQ) is apromising approach without the original datasets under privacy-sensitive andconfidential scenarios. However, current DFQ solutions degrade accuracy, needsynthetic data to calibrate networks, and are time-consuming and costly. Thispaper proposes an on-the-fly DFQ framework with sub-second quantization time,called SQuant, which can quantize networks on inference-only devices with lowcomputation and memory requirements. With the theoretical analysis of thesecond-order information of DNN task loss, we decompose and approximate theHessian-based optimization objective into three diagonal sub-items, which havedifferent areas corresponding to three dimensions of weight tensor:element-wise, kernel-wise, and output channel-wise. Then, we progressivelycompose sub-items and propose a novel data-free optimization objective in thediscrete domain, minimizing Constrained Absolute Sum of Error (or CASE inshort), which surprisingly does not need any dataset and is even not aware ofnetwork architecture. We also design an efficient algorithm withoutback-propagation to further reduce the computation complexity of the objectivesolver. Finally, without fine-tuning and synthetic datasets, SQuant acceleratesthe data-free quantization process to a sub-second level with >30% accuracyimprovement over the existing data-free post-training quantization works, withthe evaluated models under 4-bit quantization. We have open-sourced the SQuantframework at https://github.com/clevercool/SQuant.

SQuant: On-the-Fly Data-Free Quantization via Diagonal Hessian Approximation