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

Quantized neural network (NN) with a reduced bit precision is an effectivesolution to reduces the computational and memory resource requirements andplays a vital role in machine learning. However, it is still challenging toavoid the significant accuracy degradation due to its numerical approximationand lower redundancy. In this paper, a novel robustness-aware 2-bitquantization scheme is proposed for NN base on binary NN and generativeadversarial network(GAN), witch improves the performance by enriching theinformation of binary NN, efficiently extract the structural information andconsidering the robustness of the quantized NN. Specifically, using shiftaddition operation to replace the multiply-accumulate in the quantizationprocess witch can effectively speed the NN. Meanwhile, a structural lossbetween the original NN and quantized NN is proposed to such that thestructural information of data is preserved after quantization. The structuralinformation learned from NN not only plays an important role in improving theperformance but also allows for further fine tuning of the quantization networkby applying the Lipschitz constraint to the structural loss. In addition, wealso for the first time take the robustness of the quantized NN intoconsideration and propose a non-sensitive perturbation loss function byintroducing an extraneous term of spectral norm. The experiments are conductedon CIFAR-10 and ImageNet datasets with popular NN( such as MoblieNetV2,SqueezeNet, ResNet20, etc). The experimental results show that the proposedalgorithm is more competitive under 2-bit-precision than the state-of-the-artquantization methods. Meanwhile, the experimental results also demonstrate thatthe proposed method is robust under the FGSM adversarial samples attack.

Robustness-aware 2-bit quantization with real-time performance for neural network