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

Efficient model inference is an important and practical issue in thedeployment of deep neural network on resource constraint platforms. Networkquantization addresses this problem effectively by leveraging low-bitrepresentation and arithmetic that could be conducted on dedicated embeddedsystems. In the previous works, the parameter bitwidth is set homogeneously andthere is a trade-off between superior performance and aggressive compression.Actually the stacked network layers, which are generally regarded ashierarchical feature extractors, contribute diversely to the overallperformance. For a well-trained neural network, the feature distributions ofdifferent categories differentiate gradually as the network propagates forward.Hence the capability requirement on the subsequent feature extractors isreduced. It indicates that the neurons in posterior layers could be assignedwith lower bitwidth for quantized neural networks. Based on this observation, asimple but effective mixed-precision quantized neural network withprogressively ecreasing bitwidth is proposed to improve the trade-off betweenaccuracy and compression. Extensive experiments on typical networkarchitectures and benchmark datasets demonstrate that the proposed method couldachieve better or comparable results while reducing the memory space forquantized parameters by more than 30\% in comparison with the homogeneouscounterparts. In addition, the results also demonstrate that thehigher-precision bottom layers could boost the 1-bit network performanceappreciably due to a better preservation of the original image informationwhile the lower-precision posterior layers contribute to the regularization of$k-$bit networks.

Mixed-Precision Quantized Neural Network with Progressively Decreasing Bitwidth For Image Classification and Object Detection