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) has gained significant traction inautomating the design of neural networks. To reduce the time cost,differentiable architecture search (DAS) transforms the traditional paradigm ofdiscrete candidate sampling and evaluation into that of differentiablesuper-net optimization and discretization. However, existing DAS methods failto trade off between model performance and model size. They either only conductcoarse-grained operation-level search, which results in redundant modelparameters, or restrictively explore fine-grained filter-level and weight-levelunits with pre-defined remaining ratios, suffering from excessive pruningproblem. Additionally, these methods compromise search quality to save memoryduring the search process. To tackle these issues, we introducemulti-granularity architecture search (MGAS), a unified framework which aims todiscover both effective and efficient neural networks by comprehensively yetmemory-efficiently exploring the multi-granularity search space. Specifically,we improve the existing DAS methods in two aspects. First, we balance the modelunit numbers at different granularity levels with adaptive pruning. We learndiscretization functions specific to each granularity level to adaptivelydetermine the unit remaining ratio according to the evolving architecture.Second, we reduce the memory consumption without degrading the search qualityusing multi-stage search. We break down the super-net optimization anddiscretization into multiple sub-net stages, and perform progressivere-evaluation to allow for re-pruning and regrowing of previous units duringsubsequent stages, compensating for potential bias. Extensive experiments onCIFAR-10, CIFAR-100 and ImageNet demonstrate that MGAS outperforms otherstate-of-the-art methods in achieving a better trade-off between modelperformance and model size.

MGAS: Multi-Granularity Architecture Search for Trade-Off Between Model Effectiveness and Efficiency