Efficient Generation of Hidden Outliers for Improved Outlier Detection
ACM Transactions on Knowledge Discovery from Data (TKDD), 18(9)
We studied hidden outliers: points that look ordinary in the full space but stand out in some subspace. We proved that one always lies between an inlier and a full-space outlier, and turned that into BISECT, a generator that reduces detection error by up to $3\times$ in the downstream task of outlier detection with self-supervision.
Cite
@ARTICLE{cribeiro24hidden,
AUTHOR = {Jose Cribeiro-Ramallo and Vadim Arzamasov and Klemens B{\"o}hm},
TITLE = {Efficient Generation of Hidden Outliers for Improved Outlier Detection},
JOURNAL = {ACM Transactions on Knowledge Discovery from Data},
YEAR = {2024},
volume = {18},
number = {9},
pages = {1--21},
}