TY - JOUR
T1 - Asymmetric Error Control Under Imperfect Supervision
T2 - A Label-Noise-Adjusted Neyman–Pearson Umbrella Algorithm
AU - Yao, Shunan
AU - Rava, Bradley
AU - Tong, Xin
AU - James, Gareth
N1 - Publisher Copyright:
© 2022 American Statistical Association.
PY - 2023/7/3
Y1 - 2023/7/3
N2 - Label noise in data has long been an important problem in supervised learning applications as it affects the effectiveness of many widely used classification methods. Recently, important real-world applications, such as medical diagnosis and cybersecurity, have generated renewed interest in the Neyman–Pearson (NP) classification paradigm, which constrains the more severe type of error (e.g., the Type I error) under a preferred level while minimizing the other (e.g., the Type II error). However, there has been little research on the NP paradigm under label noise. It is somewhat surprising that even when common NP classifiers ignore the label noise in the training stage, they are still able to control the Type I error with high probability. However, the price they pay is excessive conservativeness of the Type I error and hence a significant drop in power (i.e., 1 - Type II error). Assuming that domain experts provide lower bounds on the corruption severity, we propose the first theory-backed algorithm that adapts most state-of-the-art classification methods to the training label noise under the NP paradigm. The resulting classifiers not only control the Type I error with high probability under the desired level but also improve power.
AB - Label noise in data has long been an important problem in supervised learning applications as it affects the effectiveness of many widely used classification methods. Recently, important real-world applications, such as medical diagnosis and cybersecurity, have generated renewed interest in the Neyman–Pearson (NP) classification paradigm, which constrains the more severe type of error (e.g., the Type I error) under a preferred level while minimizing the other (e.g., the Type II error). However, there has been little research on the NP paradigm under label noise. It is somewhat surprising that even when common NP classifiers ignore the label noise in the training stage, they are still able to control the Type I error with high probability. However, the price they pay is excessive conservativeness of the Type I error and hence a significant drop in power (i.e., 1 - Type II error). Assuming that domain experts provide lower bounds on the corruption severity, we propose the first theory-backed algorithm that adapts most state-of-the-art classification methods to the training label noise under the NP paradigm. The resulting classifiers not only control the Type I error with high probability under the desired level but also improve power.
KW - Classification
KW - Label noise
KW - Neyman–Pearson (NP) paradigm
KW - Type I error
KW - Umbrella algorithm
UR - https://www.ingentaconnect.com/content/tandf/uasa20/2023/00000118/00000543/art00033
UR - http://www.scopus.com/inward/record.url?scp=85124149128&partnerID=8YFLogxK
U2 - 10.1080/01621459.2021.2016423
DO - 10.1080/01621459.2021.2016423
M3 - Journal article
AN - SCOPUS:85124149128
SN - 0162-1459
VL - 118
SP - 1824
EP - 1836
JO - Journal of the American Statistical Association
JF - Journal of the American Statistical Association
IS - 543
ER -