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/

Zendy Plus

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

Zendy Tools

Zendy Open

One-shot Neural Architecture Search (NAS) has been widely used to discoverarchitectures due to its efficiency. However, previous studies reveal thatone-shot performance estimations of architectures might not be well correlatedwith their performances in stand-alone training because of the excessivesharing of operation parameters (i.e., large sharing extent) betweenarchitectures. Thus, recent methods construct even more over-parameterizedsupernets to reduce the sharing extent. But these improved methods introduce alarge number of extra parameters and thus cause an undesirable trade-offbetween the training costs and the ranking quality. To alleviate the aboveissues, we propose to apply Curriculum Learning On Sharing Extent (CLOSE) totrain the supernet both efficiently and effectively. Specifically, we train thesupernet with a large sharing extent (an easier curriculum) at the beginningand gradually decrease the sharing extent of the supernet (a hardercurriculum). To support this training strategy, we design a novel supernet(CLOSENet) that decouples the parameters from operations to realize a flexiblesharing scheme and adjustable sharing extent. Extensive experiments demonstratethat CLOSE can obtain a better ranking quality across different computationalbudget constraints than other one-shot supernets, and is able to discoversuperior architectures when combined with various search strategies. Code isavailable at https://github.com/walkerning/aw_nas.

CLOSE: Curriculum Learning On the Sharing Extent Towards Better One-shot NAS