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

This paper provides a novel parsimonious yet efficient design for zero-shotlearning (ZSL), dubbed ParsNets, where we are interested in learning acomposition of on-device friendly linear networks, each with orthogonality andlow-rankness properties, to achieve equivalent or even better performanceagainst existing deep models. Concretely, we first refactor the core module ofZSL, i.e., visual-semantics mapping function, into several base linear networksthat correspond to diverse components of the semantic space, where the complexnonlinearity can be collapsed into simple local linearities. Then, tofacilitate the generalization of local linearities, we construct a maximalmargin geometry on the learned features by enforcing low-rank constraints onintra-class samples and high-rank constraints on inter-class samples, resultingin orthogonal subspaces for different classes and each subspace lies on acompact manifold. To enhance the model's adaptability and counterbalanceover/under-fittings in ZSL, a set of sample-wise indicators is employed toselect a sparse subset from these base linear networks to form a compositesemantic predictor for each sample. Notably, maximal margin geometry canguarantee the diversity of features, and meanwhile, local linearities guaranteeefficiency. Thus, our ParsNets can generalize better to unseen classes and canbe deployed flexibly on resource-constrained devices. Theoretical explanationsand extensive experiments are conducted to verify the effectiveness of theproposed method.

ParsNets: A Parsimonious Orthogonal and Low-Rank Linear Networks for Zero-Shot Learning