Incomplete label distribution learning based on supervised neighborhood information
- PDF / 1,468,521 Bytes
- 11 Pages / 595.276 x 790.866 pts Page_size
- 55 Downloads / 194 Views
ORIGINAL ARTICLE
Incomplete label distribution learning based on supervised neighborhood information Xue‑Qiang Zeng1 · Su‑Fen Chen2 · Run Xiang3 · Guo‑Zheng Li4 · Xue‑Feng Fu2 Received: 27 September 2018 / Accepted: 23 April 2019 © Springer-Verlag GmbH Germany, part of Springer Nature 2019
Abstract Label distribution learning (LDL) assumes labels are associated with each instance to some degree and tries to model the relationship between labels and instances. LDL has achieved great success in many applications, but most existing LDL methods are designed for data with complete annotation information. However, in reality, supervised information often be incomplete due to the huge costs of data collection. In this paper, we propose a novel incomplete label distribution learning method based on supervised neighborhood information (IncomLDL-SNI). The proposed method uses partial least squares to project the original data into a supervised feature space where instances with similar labels are likely to be projected together. Then, IncomLDL-SNI utilizes the Euclidean distance to find the nearest neighbors for target samples in the supervised feature space and recovers the missing annotations from the neighborhood label Information. Extensive experiments on various data sets validate the effectiveness of our proposal. Keywords Label distribution learning · Incomplete annotation · Partial least squares · Supervised neighborhood information
1 Introduction
* Guo‑Zheng Li [email protected] Xue‑Qiang Zeng [email protected] Su‑Fen Chen [email protected] Run Xiang [email protected] Xue‑Feng Fu [email protected] 1
School of Computer and Information Engineering, Jiangxi Normal University, 99 Ziyang Road, Nanchang 330022, China
2
School of Information Engineering, Nanchang Institute of Technology, 289 Tianxiang Road, Nanchang 330099, China
3
Information Engineering School, Nanchang University, 999 Xuefu Road, Nanchang 330031, China
4
Data Center of Traditional Chinese Medicine, China Academy of Chinese Medical Science, 16 DongZhiMenNeiNanXiaoJie, Beijing 100700, China
Classical machine learning tasks assume that one instance is associated with one label or some labels, where the relationship between label and instance is hard assignment. However, in many practical applications, labels may be associated with an instance to some degree, and it is difficult to learn this kind of relationship by traditional machine learning methods. Label distribution learning (LDL) [1] uses soft labels rather than a single label or a set of labels to annotate an instance, and then learning a mapping from an instance to all labels. LDL has been successfully used in many practical applications in recent years, such as facial age estimation [2, 3], crowd opinion prediction [4], facial expression recognition [5], head pose estimation [6] and so on [7–15]. Most existing LDL methods are designed for data with complete supervised information. Nevertheless, in reality, supervised information of data is often incomplete. Annota
Data Loading...
