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

Designing effective architectures is one of the key factors behind thesuccess of deep neural networks. Existing deep architectures are eithermanually designed or automatically searched by some Neural Architecture Search(NAS) methods. However, even a well-searched architecture may still containmany non-significant or redundant modules or operations (e.g., convolution orpooling), which may not only incur substantial memory consumption andcomputation cost but also deteriorate the performance. Thus, it is necessary tooptimize the operations inside an architecture to improve the performancewithout introducing extra computation cost. Unfortunately, such a constrainedoptimization problem is NP-hard. To make the problem feasible, we cast theoptimization problem into a Markov decision process (MDP) and seek to learn aNeural Architecture Transformer (NAT) to replace the redundant operations withthe more computationally efficient ones (e.g., skip connection or directlyremoving the connection). Based on MDP, we learn NAT by exploitingreinforcement learning to obtain the optimization policies w.r.t. differentarchitectures. To verify the effectiveness of the proposed strategies, we applyNAT on both hand-crafted architectures and NAS based architectures. Extensiveexperiments on two benchmark datasets, i.e., CIFAR-10 and ImageNet, demonstratethat the transformed architecture by NAT significantly outperforms both itsoriginal form and those architectures optimized by existing methods.

NAT: Neural Architecture Transformer for Accurate and Compact Architectures