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

Corrupted labels and class imbalance are commonly encountered in practicallycollected training data, which easily leads to over-fitting of deep neuralnetworks (DNNs). Existing approaches alleviate these issues by adopting asample re-weighting strategy, which is to re-weight sample by designingweighting function. However, it is only applicable for training data containingonly either one type of data biases. In practice, however, biased samples withcorrupted labels and of tailed classes commonly co-exist in training data. Howto handle them simultaneously is a key but under-explored problem. In thispaper, we find that these two types of biased samples, though have similartransient loss, have distinguishable trend and characteristics in loss curves,which could provide valuable priors for sample weight assignment. Motivated bythis, we delve into the loss curves and propose a novel probe-and-allocatetraining strategy: In the probing stage, we train the network on the wholebiased training data without intervention, and record the loss curve of eachsample as an additional attribute; In the allocating stage, we feed theresulting attribute to a newly designed curve-perception network, namedCurveNet, to learn to identify the bias type of each sample and assign properweights through meta-learning adaptively. The training speed of meta learningalso blocks its application. To solve it, we propose a method named skip layermeta optimization (SLMO) to accelerate training speed by skipping the bottomlayers. Extensive synthetic and real experiments well validate the proposedmethod, which achieves state-of-the-art performance on multiple challengingbenchmarks.

Delving into Sample Loss Curve to Embrace Noisy and Imbalanced Data