TY - GEN
T1 - Label-Assisted Memory Autoencoder for Unsupervised Out-of-Distribution Detection
AU - ZHANG, Shuyi
AU - PAN, Chao
AU - SONG, Liyan
AU - WU, Xiaoyu
AU - HU, Zheng
AU - PEI, Ke
AU - TINO, Peter
AU - YAO, Xin
N1 - This work was supported by National Natural Science Foundation of China (Grant No. 62002148), the Guangdong Provincial Key Laboratory (Grant No. 2020B121201001), the Program for Guangdong Introducing Innovative and Entrepreneurial Teams (Grant No. 2017ZT07X386), Shenzhen Science and Technology Program (Grant No. KQTD2016112514355531), Shenzhen Fundamental Research Program (Grant No. JCYJ20190809121403553), Research Institute of Trustworthy Autonomous Systems, and Huawei.
PY - 2021
Y1 - 2021
N2 - Out-of-Distribution (OoD) detectors based on AutoEncoder (AE) rely on an underlying assumption that an AE network cannot reconstruct OoD data as good as in-distribution (ID) data when it is constructed based on ID data only. However, this assumption may be violated in practice, resulting in a degradation in detection performance. Therefore, alleviating the factors violating this assumption can potentially improve the robustness of OoD performance. Our empirical studies also show that image complexity can be another factor hindering detection performance for AE-based detectors. To cater for these issues, we propose two OoD detectors LAMAE and LAMAE+. Both can be trained without the availability of any OoD-related data. The key idea is to regularize the AE network architecture with a classifier and a label-assisted memory to confine the reconstruction of OoD data while retaining the reconstruction ability for ID data. We also adjust the reconstruction error by taking image complexity into consideration. Experimental studies show that the proposed OoD detectors can perform well on a wider range of OoD scenarios. © 2021, Springer Nature Switzerland AG.
AB - Out-of-Distribution (OoD) detectors based on AutoEncoder (AE) rely on an underlying assumption that an AE network cannot reconstruct OoD data as good as in-distribution (ID) data when it is constructed based on ID data only. However, this assumption may be violated in practice, resulting in a degradation in detection performance. Therefore, alleviating the factors violating this assumption can potentially improve the robustness of OoD performance. Our empirical studies also show that image complexity can be another factor hindering detection performance for AE-based detectors. To cater for these issues, we propose two OoD detectors LAMAE and LAMAE+. Both can be trained without the availability of any OoD-related data. The key idea is to regularize the AE network architecture with a classifier and a label-assisted memory to confine the reconstruction of OoD data while retaining the reconstruction ability for ID data. We also adjust the reconstruction error by taking image complexity into consideration. Experimental studies show that the proposed OoD detectors can perform well on a wider range of OoD scenarios. © 2021, Springer Nature Switzerland AG.
UR - http://www.scopus.com/inward/record.url?scp=85115709907&partnerID=8YFLogxK
U2 - 10.1007/978-3-030-86523-8_48
DO - 10.1007/978-3-030-86523-8_48
M3 - Conference paper (refereed)
SN - 9783030865221
T3 - Lecture Notes in Computer Science
SP - 795
EP - 810
BT - Machine Learning and Knowledge Discovery in Databases. Research Track : European Conference, ECML PKDD 2021, Bilbao, Spain, September 13–17, 2021, Proceedings, Part III
A2 - OLIVER, Nuria
A2 - PÉREZ-CRUZ, Fernando
A2 - KRAMER, Stefan
A2 - READ, Jesse
A2 - LOZANO, Jose A.
PB - Springer Science and Business Media Deutschland GmbH
T2 - European Conference on Machine Learning and Principles and Practice of Knowledge Discovery in Databases 2021
Y2 - 13 September 2021 through 17 September 2021
ER -