Cognitive concept learning via granular computing for big data Jinhai Li 1 , Chenchen Huang 1 , Weihua Xu 2 , Yuhua Qian 3 , Wenqi Liu 1 1 Faculty of Science, Kunming University of Science and Technology, Kunming 650500, PR China 2 School of Mathematics and Statistics, Chongqing University of Technology, Chongqing 400054, PR China 3 Key Laboratory of Computational Intelligence and Chinese Information Processing of Ministry of Education, School of Computer and Information Technology, Shanxi University, Taiyuan 030006, PR China E-MAIL: jhlixjtu@163.com, huangchenchen64@163.com, chxuwh@gmail.com, jinchengqyh@126.com Abstract: In this paper, analysis of time complexity of the exist- ing granular-computing-based cognitive concept learning meth- ods is made by numerical experiments. Moreover, some technical problems are addressed when the existing cogni- tive concept learning methods are applied to big data. Keywords: Concept learning; Cognitive computing; Granular computing; Complexity; Big data 1. Introduction Cognitive concept learning has become a hot topic in recent years [1, 2, 3]. It can be regarded as an effective way of learning concepts from the perspective of cognition [4, 5, 6]. Note that here the notion of a concept is not limited to the one proposed by Wille [7], but all of the ones [8, 9, 10, 11, 12, 13, 14, 15, 16] generated by constructive and/or axiomatic methods. General- ly speaking, the study of cognitive concept learning consists of analyzing cognitive mechanism of forming concepts, building cognitive computing system, and designing cognitive process- es. It deserved to be mentioned that sets of axioms were usu- ally used to analyze cognitive mechanism of forming concepts [17, 18], cognitive operators and their updating were common- ly employed to build cognitive computing system [2], and con- structive and approximation ideas were often taken to design cognitive processes [1, 2, 3]. Considering that cognitive concept learning will become computationally expensive when it learns the whole concep- t structure, granular computing [19] was integrated into it for improving learning efficiency [2], since it is sufficient to de- termine the whole concept structure through granular concepts equipped with some reasonable operators [20]. Moreover, even though one does not need to learn the whole concept structure, information granules [21, 22], the basic notion in the theory of granular computing, are quite necessary for interpreting mean- ings of real-world concrete entities and semantics of abstract subjects [4, 5, 6]. Therefore, cognitive concept learning via granular computing is a good way of improving learning effi- ciency and interpreting the learnt knowledge. To the best of our knowledge, there have been some studies on the issue of cognitive concept learning via granular comput- ing. For example, reference [1] put forward a static cognitive computing system for learning concepts from the given infor- mation, which was also extended to the case of fuzzy informa- tion [23]. Considering that a static cognitive computing sys- tem will become unapplicable for dealing with dynamic infor- mation, reference [2] designed a dynamic cognitive computing system to solve this problem. In addition, since cognitive abili- ties of human brain such as learning, memorizing, recalling and updating of information can effectively be performed via con- cepts, reference [3] investigated how to learn, bi-directionally recall and update the information by cognitive concept learn- ing. Although some attempts have been made to study cogni- tive concept learning via granular computing, there are stil- l many challenging problems. For example, the existing granular-computing-based cognitive concept learning methods have been tested only on toy examples or small data sets, and it is not yet known whether they can be effective on large data sets. Another problem is that the existing cognitive concep- t learning methods via granular computing were designed for common data, which means that they may become unapplica- ble for big data, since big data starts with large-volume, het- erogeneous, autonomous sources with distributed and decen- tralized control [24, 25, 26, 27]. Motivated by these problems, this study makes a complexity study of the existing granular- 289 2015 IEEE 978-1-4673-7220-6/15/$31.00 © Proceedings of the 2015 International Conference on Machine Learning and Cybernetics, Guangzhou, 12-15 July, 2015