TY - GEN
T1 - Top-Rank Learning Robust to Outliers
AU - Zheng, Yan
AU - Suehiro, Daiki
AU - Uchida, Seiichi
N1 - Funding Information:
This work was supported by MEXT-Japan (Grant No. J17H06100), and JST, ACT-X, Japan (Grant No. JPMJAX200G), and China Scholarship Council (Grant No. 201806330079).
Publisher Copyright:
© 2021, Springer Nature Switzerland AG.
PY - 2021
Y1 - 2021
N2 - Top-rank learning aims to maximize the number of absolute top samples, which are “doubtlessly positive” samples and very useful for the real applications that require reliable positive samples. However, top-rank learning is very sensitive to outliers of the negative class. This paper proposes a robust top-rank learning algorithm with an unsupervised outlier estimation technique called local outlier factor (LoF). Introduction of LoF can weaken the effect of the negative outliers and thus increase the stability of the learned ranking function. Moreover, we combine robust top-rank learning with representation learning by a deep neural network (DNN). Experiments on artificial datasets and a medical image dataset demonstrate the robustness of the proposed method to outliers.
AB - Top-rank learning aims to maximize the number of absolute top samples, which are “doubtlessly positive” samples and very useful for the real applications that require reliable positive samples. However, top-rank learning is very sensitive to outliers of the negative class. This paper proposes a robust top-rank learning algorithm with an unsupervised outlier estimation technique called local outlier factor (LoF). Introduction of LoF can weaken the effect of the negative outliers and thus increase the stability of the learned ranking function. Moreover, we combine robust top-rank learning with representation learning by a deep neural network (DNN). Experiments on artificial datasets and a medical image dataset demonstrate the robustness of the proposed method to outliers.
UR - http://www.scopus.com/inward/record.url?scp=85121907295&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=85121907295&partnerID=8YFLogxK
U2 - 10.1007/978-3-030-92238-2_50
DO - 10.1007/978-3-030-92238-2_50
M3 - Conference contribution
AN - SCOPUS:85121907295
SN - 9783030922375
T3 - Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
SP - 608
EP - 619
BT - Neural Information Processing - 28th International Conference, ICONIP 2021, Proceedings
A2 - Mantoro, Teddy
A2 - Lee, Minho
A2 - Ayu, Media Anugerah
A2 - Wong, Kok Wai
A2 - Hidayanto, Achmad Nizar
PB - Springer Science and Business Media Deutschland GmbH
T2 - 28th International Conference on Neural Information Processing, ICONIP 2021
Y2 - 8 December 2021 through 12 December 2021
ER -