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

Zero-shot hyperparameter optimization (HPO) is a simple yet effective use oftransfer learning for constructing a small list of hyperparameter (HP)configurations that complement each other. That is to say, for any givendataset, at least one of them is expected to perform well. Current techniquesfor obtaining this list are computationally expensive as they rely on runningtraining jobs on a diverse collection of datasets and a large collection ofrandomly drawn HPs. This cost is especially problematic in environments wherethe space of HPs is regularly changing due to new algorithm versions, orchanging architectures of deep networks. We provide an overview of availableapproaches and introduce two novel techniques to handle the problem. The firstis based on a surrogate model and adaptively chooses pairs of dataset,configuration to query. The second, for settings where finding, tuning andtesting a surrogate model is problematic, is a multi-fidelity techniquecombining HyperBand with submodular optimization. We benchmark our methodsexperimentally on five tasks (XGBoost, LightGBM, CatBoost, MLP and AutoML) andshow significant improvement in accuracy compared to standard zero-shot HPOwith the same training budget. In addition to contributing new algorithms, weprovide an extensive study of the zero-shot HPO technique resulting in (1)default hyper-parameters for popular algorithms that would benefit thecommunity using them, (2) massive lookup tables to further the research ofhyper-parameter tuning.

Practical and sample efficient zero-shot HPO