Autoencoding Under Normalization Constraints
- Authors
- Yoon, Sangwoong; Noh, Yung-Kyun; Park, Frank C.
- Issue Date
- Jul-2021
- Publisher
- JMLR-JOURNAL MACHINE LEARNING RESEARCH
- Citation
- INTERNATIONAL CONFERENCE ON MACHINE LEARNING, VOL 139, v.139, pp.12087 - 12097
- Indexed
- SCOPUS
- Journal Title
- INTERNATIONAL CONFERENCE ON MACHINE LEARNING, VOL 139
- Volume
- 139
- Start Page
- 12087
- End Page
- 12097
- URI
- https://scholarworks.bwise.kr/hanyang/handle/2021.sw.hanyang/141471
- DOI
- 10.48550/arXiv.2105.05735
- ISSN
- 2640-3498
- Abstract
- Likelihood is a standard estimate for outlier detection. The specific role of the normalization constraint is to ensure that the out-of-distribution (OOD) regime has a small likelihood when samples are learned using maximum likelihood. Because autoencoders do not possess such a process of normalization, they often fail to recognize outliers even when they are obviously OOD. We propose the Normalized Autoencoder (NAE), a normalized pro-babilistic model constructed from an autoencoder. The probability density of NAE is defined using the reconstruction error of an autoencoder, which is differently defined in the conventional energy-based model. In our model, normalization is enforced by suppressing the reconstruction of negative samples, significantly improving the outlier detection performance. Our experimental results confirm the efficacy of NAE, both in detecting outliers and in generating in-distribution samples.
- Files in This Item
-
Go to Link
- Appears in
Collections - 서울 공과대학 > 서울 컴퓨터소프트웨어학부 > 1. Journal Articles
![qrcode](https://api.qrserver.com/v1/create-qr-code/?size=55x55&data=https://scholarworks.bwise.kr/hanyang/handle/2021.sw.hanyang/141471)
Items in ScholarWorks are protected by copyright, with all rights reserved, unless otherwise indicated.