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

Feature embeddings are one of the most essential steps when training deeplearning based Click-Through Rate prediction models, which map high-dimensionalsparse features to dense embedding vectors. Classic human-crafted embeddingsize selection methods are shown to be "sub-optimal" in terms of the trade-offbetween memory usage and model capacity. The trending methods in NeuralArchitecture Search (NAS) have demonstrated their efficiency to search forembedding sizes. However, most existing NAS-based works suffer from expensivecomputational costs, the curse of dimensionality of the search space, and thediscrepancy between continuous search space and discrete candidate space. Otherworks that prune embeddings in an unstructured manner fail to reduce thecomputational costs explicitly. In this paper, to address those limitations, wepropose a novel strategy that searches for the optimal mixed-dimensionembedding scheme by structurally pruning a super-net via Hard Auxiliary Mask.Our method aims to directly search candidate models in the discrete space usinga simple and efficient gradient-based method. Furthermore, we introduceorthogonal regularity on embedding tables to reduce correlations withinembedding columns and enhance representation capacity. Extensive experimentsdemonstrate it can effectively remove redundant embedding dimensions withoutgreat performance loss.

Field-wise Embedding Size Search via Structural Hard Auxiliary Mask Pruning for Click-Through Rate Prediction