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

Differentiable neural architecture search (DARTS) has gained much success indiscovering flexible and diverse cell types. To reduce the evaluation gap, thesupernet is expected to have identical layers with the target network. However,even for this consistent search, the searched cells often suffer from poorperformance, especially for the supernet with fewer layers, as current DARTSmethods are prone to wide and shallow cells, and this topology collapse inducessub-optimal searched cells. In this paper, we alleviate this issue by endowingthe cells with explicit stretchability, so the search can be directlyimplemented on our stretchable cells for both operation type and topologysimultaneously. Concretely, we introduce a set of topological variables and acombinatorial probabilistic distribution to explicitly model the targettopology. With more diverse and complex topologies, our method adapts well forvarious layer numbers. Extensive experiments on CIFAR-10 and ImageNet show thatour stretchable cells obtain better performance with fewer layers andparameters. For example, our method can improve DARTS by 0.28\% accuracy onCIFAR-10 dataset with 45\% parameters reduced or 2.9\% with similar FLOPs onImageNet dataset.

Stretchable Cells Help DARTS Search Better