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

Most existing works in few-shot learning rely on meta-learning the network ona large base dataset which is typically from the same domain as the targetdataset. We tackle the problem of cross-domain few-shot learning where there isa large shift between the base and target domain. The problem of cross-domainfew-shot recognition with unlabeled target data is largely unaddressed in theliterature. STARTUP was the first method that tackles this problem usingself-training. However, it uses a fixed teacher pretrained on a labeled basedataset to create soft labels for the unlabeled target samples. As the basedataset and unlabeled dataset are from different domains, projecting the targetimages in the class-domain of the base dataset with a fixed pretrained modelmight be sub-optimal. We propose a simple dynamic distillation-based approachto facilitate unlabeled images from the novel/base dataset. We imposeconsistency regularization by calculating predictions from the weakly-augmentedversions of the unlabeled images from a teacher network and matching it withthe strongly augmented versions of the same images from a student network. Theparameters of the teacher network are updated as exponential moving average ofthe parameters of the student network. We show that the proposed network learnsrepresentation that can be easily adapted to the target domain even though ithas not been trained with target-specific classes during the pretraining phase.Our model outperforms the current state-of-the art method by 4.4% for 1-shotand 3.6% for 5-shot classification in the BSCD-FSL benchmark, and also showscompetitive performance on traditional in-domain few-shot learning task.

Dynamic Distillation Network for Cross-Domain Few-Shot Recognition with Unlabeled Data