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 Tools annual

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 Tools monthly

Global: Zendy Plus monthly

Presents the access of premium content as premium feature

Premium Content

Global: Zendy Plus annual

Global: Zendy Free Plan

Neural architecture search (NAS) proves to be among the effective approachesfor many tasks by generating an application-adaptive neural architecture, whichis still challenged by high computational cost and memory consumption. At thesame time, 1-bit convolutional neural networks (CNNs) with binary weights andactivations show their potential for resource-limited embedded devices. Onenatural approach is to use 1-bit CNNs to reduce the computation and memory costof NAS by taking advantage of the strengths of each in a unified framework,while searching the 1-bit CNNs is more challenging due to the more complicatedprocesses involved. In this paper, we introduce Discrepant Child-Parent NeuralArchitecture Search (DCP-NAS) to efficiently search 1-bit CNNs, based on a newframework of searching the 1-bit model (Child) under the supervision of areal-valued model (Parent). Particularly, we first utilize a Parent model tocalculate a tangent direction, based on which the tangent propagation method isintroduced to search the optimized 1-bit Child. We further observe a couplingrelationship between the weights and architecture parameters existing in suchdifferentiable frameworks. To address the issue, we propose a decoupledoptimization method to search an optimized architecture. Extensive experimentsdemonstrate that our DCP-NAS achieves much better results than prior arts onboth CIFAR-10 and ImageNet datasets. In particular, the backbones achieved byour DCP-NAS achieve strong generalization performance on personre-identification and object detection.

DCP-NAS: Discrepant Child-Parent Neural Architecture Search for 1-bit CNNs