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

Rare categories abound in a number of real-world networks and play a pivotalrole in a variety of high-stakes applications, including financial frauddetection, network intrusion detection, and rare disease diagnosis. Rarecategory analysis (RCA) refers to the task of detecting, characterizing, andcomprehending the behaviors of minority classes in a highly-imbalanced datadistribution. While the vast majority of existing work on RCA has focused onimproving the prediction performance, a few fundamental research questionsheretofore have received little attention and are less explored: How confidentor uncertain is a prediction model in rare category analysis? How can wequantify the uncertainty in the learning process and enable reliable rarecategory analysis?  To answer these questions, we start by investigating miscalibration inexisting RCA methods. Empirical results reveal that state-of-the-art RCAmethods are mainly over-confident in predicting minority classes andunder-confident in predicting majority classes. Motivated by the observation,we propose a novel individual calibration framework, named CALIRARE, foralleviating the unique challenges of RCA, thus enabling reliable rare categoryanalysis. In particular, to quantify the uncertainties in RCA, we develop anode-level uncertainty quantification algorithm to model the overlappingsupport regions with high uncertainty; to handle the rarity of minority classesin miscalibration calculation, we generalize the distribution-based calibrationmetric to the instance level and propose the first individual calibrationmeasurement on graphs named Expected Individual Calibration Error (EICE). Weperform extensive experimental evaluations on real-world datasets, includingrare category characterization and model calibration tasks, which demonstratethe significance of our proposed framework.

Towards Reliable Rare Category Analysis on Graphs via Individual Calibration