A new multi-task learning method with universum data
- PDF / 1,798,613 Bytes
- 14 Pages / 595.224 x 790.955 pts Page_size
- 33 Downloads / 149 Views
A new multi-task learning method with universum data Yanshan Xiao1 · Jing Wen1 · Bo Liu2 Accepted: 16 September 2020 © Springer Science+Business Media, LLC, part of Springer Nature 2020
Abstract Multi-task learning (MTL) obtains a better classifier than single-task learning (STL) by sharing information between tasks within the multi-task models. Most existing multi-task learning models only focus on the data of the target tasks during training, and ignore the data of non-target tasks that may be contained in the target tasks. In this way, Universum data can be added to classifier training as prior knowledge, and these data do not belong to any indicated categories. In this paper, we address the problem of multi-task learning with Universum data, which improves utilization of non-target task data. We introduce Universum learning to make non-target task data act as prior knowledge and propose a novel multi-task support vector machine with Universum data (U-MTLSVM). Based on the characteristics of MTL, each task have corresponding Universum data to provide prior knowledge. We then utilize the Lagrange method to solve the optimization problem so as to obtain the multi-task classifiers. Then, conduct experiments to compare the performance of the proposed method with several baslines on different data sets. The experimental results demonstrate the effectiveness of the proposed methods for multi-task classification. Keywords Multi-task learning · Universum learning · SVM · Prior knowledge
1 Introduction Traditional machine learning always focuses on learning a model from one task, which can be called as single-task learning [1–3]. However, we can meet the learning case in which the data come from several domains, and each domain data have similar distribution with each other. We then need to build a model catering for the multi-task data, in which each task can help other tasks to build its predictive model. This is always called multi-task learning [4, 5]. Compared with single-task learning, multi-task learning can make better use of the information contained in related tasks to help build the model. To date, multi-task learning has been successfully used in many applications, such as in speech recognition [6], natural language processing [7], images recognition [8]. For example, in images recognition, the work in [9] uses a multitasking learning framework to Bo Liu
[email protected] 1
Department of Computer Science, Guangdong University of Technology, Guangzhou, Guangdong, China
2
Department of Automation, Guangdong University of Technology, Guangzhou, Guangdong, China
build models that take advantage of various types of tag information to represent clothing images in a more granular way. Over the years, quite a number of MTL methods are proposed, and they can be broadly grouped into the categories of SVM-based methods [10, 11], neural workbased methods [12, 13], and Bayesian based methods [14, 15]. For the SVM-based methods, each task shares a common parameters for the classifier, and builds a classifier similar
Data Loading...
