Multi-label Active Learning Based on Maximum Correntropy Criterion: Towards Robust and Discriminative Labeling
Multi-label learning is a challenging problem in computer vision field. In this paper, we propose a novel active learning approach to reduce the annotation costs greatly for multi-label classification. State-of-the-art active learning methods either annot
- PDF / 1,535,928 Bytes
- 16 Pages / 439.37 x 666.142 pts Page_size
- 56 Downloads / 194 Views
State Key Laboratory of Software Engineering, School of Computer, Wuhan University, Wuhan, China {kingmao,remoteking,zhanglefei}@whu.edu.cn 2 State Key Laboratory of Information Engineering in Surveying, Mapping and Remote Sensing, Wuhan University, Wuhan, China [email protected] 3 Department of Computing and Information Systems, University of Melbourne, Parkville, Australia [email protected] 4 QCIS and FEIT, University of Technology Sydney, Sydney, NSW 2007, Australia [email protected]
Abstract. Multi-label learning is a challenging problem in computer vision field. In this paper, we propose a novel active learning approach to reduce the annotation costs greatly for multi-label classification. State-of-the-art active learning methods either annotate all the relevant samples without diagnosing discriminative information in the labels or annotate only limited discriminative samples manually, that has weak immunity for the outlier labels. To overcome these problems, we propose a multi-label active learning method based on Maximum Correntropy Criterion (MCC) by merging uncertainty and representativeness. We use the the labels of labeled data and the prediction labels of unknown data to enhance the uncertainty and representativeness measurement by merging strategy, and use the MCC to alleviate the influence of outlier labels for discriminative labeling. Experiments on several challenging benchmark multi-label datasets show the superior performance of our proposed method to the state-of-the-art methods. Keywords: Multi-label learning Robust
1
·
Active learning
·
Correntropy
·
Introduction
Active learning has been widely used in computer visions to address the samples imbalance problem that the available labeled data is much less than the unlabeled data [18,35]. It is an iterative loop to find the most valuable samples for the oracle to label, and gradually improves the model generalization ability until the convergence condition is satisfied [39]. There are two motivations behind the design of c Springer International Publishing AG 2016 B. Leibe et al. (Eds.): ECCV 2016, Part III, LNCS 9907, pp. 453–468, 2016. DOI: 10.1007/978-3-319-46487-9 28
454
Z. Wang et al.
a practical active learning algorithm, namely, uncertainty and representativeness [8,15]. Uncertainty is to improve the models’ generalization ability and representativeness is to prevent the bias of the models. Among all the active learning based tasks, multi-label classification, which aims to assign each object with multiple labels, may be the most difficult and costly one [10,17,42]. In current research, active learning for multi-label learning has become even more important, reducing the costs of the various multi-label tasks [6,7,38,41]. State-of-the-art multi-label active learning can be classified into three categories based on the query function used to select the valuable samples. The first category relies on the labeled data to design a query function with uncertainty [25,28]. In such methods, the design of the query function ignores t
