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 is a technique to reduce the computation and memory cost of DNNmodels, which are getting increasingly large. Existing quantization solutionsuse fixed-point integer or floating-point types, which have limited benefits,as both require more bits to maintain the accuracy of original models. On theother hand, variable-length quantization uses low-bit quantization for normalvalues and high-precision for a fraction of outlier values. Even though thisline of work brings algorithmic benefits, it also introduces significanthardware overheads due to variable-length encoding and decoding.  In this work, we propose a fixed-length adaptive numerical data type calledANT to achieve low-bit quantization with tiny hardware overheads. Our data typeANT leverages two key innovations to exploit the intra-tensor and inter-tensoradaptive opportunities in DNN models. First, we propose a particular data type,flint, that combines the advantages of float and int for adapting to theimportance of different values within a tensor. Second, we propose an adaptiveframework that selects the best type for each tensor according to itsdistribution characteristics. We design a unified processing elementarchitecture for ANT and show its ease of integration with existing DNNaccelerators. Our design results in 2.8$\times$ speedup and 2.5$\times$ energyefficiency improvement over the state-of-the-art quantization accelerators.

ANT: Exploiting Adaptive Numerical Data Type for Low-bit Deep Neural Network Quantization