Search

Distance metric learning aims to learn a data-dependent similarity measure, which is widely employed in machine learning. Recently, metric learning algorithms that incorporate multiple kernel learning have shown promising outcomes for classification tasks. However, the multiple kernel learning part of the existing metric learning with multiple kernel just uses a linear combination form of different kernel functions, where each kernel shares the same weight in the entire input space, thus the potential local structure of samples located at different locations in the input space is ignored. To address the aforementioned issues, in this paper, we propose a distance metric learning approach with local multiple kernel embedding (DMLLMK) for small datasets. The weight of each kernel function in DMLLMK is assigned locally, so that there are many different values of weight in each kernel space. This local weight method enables metric learning to capture more information in the data. Our proposed DMLLMK adjusts the kernel weight by using a gating function; moreover, the kernel weight locally depends on the input data. The metric of metric learning and the parameters of the gating function are optimized simultaneously by an alternating learning process. The DMLLMK makes metric learning applicable to small datasets by constructing constraints on the set of similar pairs and dissimilar pairs such that some data are reused, and they produce different constraints on the model. In addition, regularization techniques are used to keep DMLLMK more conservative and prevent overfitting on small data. The experimental results of our proposed method when compared with other metric learning methods on the benchmark dataset show that our proposed DMLLMK is effective. [ABSTRACT FROM AUTHOR]

Concept-cognitive learning (CCL) is to make machines like human beings have the ability of summarizing and reasoning. Automatically learn and find concepts from given information clues is a research focus of CCL. The existing researches mainly focuses on the concept learning methods in classical and fuzzy formal contexts, but there are few researches on the CCL of interval-valued contexts. In view of the universality of interval values in practical applications, we study the mechanism of CCL in interval-valued formal contexts. Firstly, we propose interval-valued formal contexts and a pair of dual cognitive operators as the fundamental foundation of concept learning. Then we mine the relationship between interval-valued information granules and concepts from cognitive learning and granular computing perspective. Then we systematically study the mechanism of interval-valued CCL from the establishment of interval-valued information granules (IvIGs) and its mathematical properties, and the transformation between different information granules (IGs) and clue oriented concept learning. Moreover, three algorithms are established to automatically learn concepts from different clue information. Finally, we download eight public data sets to verify the effectiveness and feasibility of the proposed algorithms from the perspective of the size of extension of concepts, running time of concept learning algorithms and the number of concepts learned by the concept learning algorithms. The experimental comparison indicates that the proposed algorithms are effective and feasible for interval-valued CCL. [ABSTRACT FROM AUTHOR]

A causative attack which manipulates training samples to mislead learning is a common attack scenario. Current countermeasures reduce the influence of the attack to a classifier with the loss of generalization ability. Therefore, the collected samples should be analyzed carefully. Most countermeasures of current causative attack focus on data sanitization and robust classifier design. To our best knowledge, there is no work to determinate whether a given dataset is contaminated by a causative attack. In this study, we formulate a causative attack detection as a 2-class classification problem in which a sample represents a dataset quantified by data complexity measures, which describe the geometrical characteristics of data. As geometrical natures of a dataset are changed by a causative attack, we believe data complexity measures provide useful information for causative attack detection. Furthermore, a two-step secure classification model is proposed to demonstrate how the proposed causative attack detection improves the robustness of learning. Either a robust or traditional learning method is used according to the existence of causative attack. Experimental results illustrate that data complexity measures separate untainted datasets from attacked ones clearly, and confirm the promising performance of the proposed methods in terms of accuracy and robustness. The results consistently suggest that data complexity measures provide the crucial information to detect causative attack, and are useful to increase the robustness of learning. [ABSTRACT FROM AUTHOR]

As a new interdisciplinary field induced by formal concept analysis, rough set, granular computing and cognitive computing, cognitive concept learning has received a great attention in recent years. Cognitive concept learning refers to the acquisition of specific concepts through specific cognitive concept learning approaches. The processes of concept learning mainly focus on simulating human brain recognizing concepts through the modeling of brain intelligence. In this paper, we investigate the mechanism of multi-level cognitive concept learning method oriented to data sets with fuzziness by discussing the process of human cognition. Through a newly defined fuzzy focal feature set, we put forward a corresponding structure of feature-oriented multi-level cognitive concept learning method in data sets with fuzziness from a perspective of philosophical and psychological views of human cognition. To make the presented cognitive concept learning approach much easier to understand and to apply it to practice widely, we establish an algorithm to recognize fuzzy concepts and incomplete fuzzy concepts. In addition, we present a case study about how to recognize and distinguish any two different micro-expressions from an information system with quantitative description to use our proposed method and theory to solve conceptual cognition problems, and also we perform an experimental evaluation on five data sets downloaded from the University of California-Irvine databases. Compared with the existing granular computing approach to two-way learning, we obtain more concepts than the two-way learning approach, which shows the feasibility and effectiveness of our feature-oriented multi-level cognitive learning method in data sets with fuzziness. [ABSTRACT FROM AUTHOR]

Recently, formal concept analysis has become a potential direction of cognitive computing, which can describe the processes of cognitive concept learning. We establish a concept hierarchy structure based on the existing cognitive concept learning methods. However, none of these methods could obtain the following results: get the concept, recognize objects and distinguish between two different objects. In this paper, our focus is to construct an attribute-oriented multi-level cognitive concept learning method so as to improve and enhance the ability of cognitive concept learning. Firstly, the view point of human cognition is discussed from the multi-level approach, and then the mechanism of attribute-oriented cognitive concept learning is investigated. Through some defined special attributes, we propose a corresponding structure of attribute-oriented multi-level cognitive concept learning from an interdisciplinary viewpoint. It is a combination of philosophy and psychology of human cognition. Moreover, to make the presented attribute-oriented multi-level method easier to understand and apply in practice, an algorithm of cognitive concept learning is established. Furthermore, a case study about how to recognize the real-world animals is studied to use the proposed method and theory. Finally, in order to solve conceptual cognition problems, we perform an experimental evaluation on five data sets downloaded from the University of California-Irvine (UCI) databases. And then we provide a comparative analysis with the existing g r a n u l a r c o m p u t i n g a p p r o a c h t o t w o - w a y l e a r n i n g [44] and the three- w a y c o g n i t i v e c o n c e p t l e a r n i n g v i a m u l t i -granularity [9]. We obtain more number of concepts than t h e t w o - w a y l e a r n i n g a n d t h e t h r e e - w a y c o g n i t i v e c o n c e p t l e a r n i n g a p p r o a c h e s , which shows the feasibility and effectiveness of our attribute-oriented multi-level cognitive learning method. [ABSTRACT FROM AUTHOR]

Extreme learning machine (ELM) is a non-iterative algorithm for training single-hidden layer feed-forward networks, whose training speed is much faster than those of conventional neural networks. However, its objective is only to minimize the empirical risk, which may cause overfitting easily. To overcome this defect, this paper proposes a novel algorithm named LGE2LM based on the localized generalization error model which provides an upper bound of generalization error through adopting the stochastic sensitivity. LGE2LM aims to improve the generalization capability of ELM by using the regularization technique which makes a trade-off between the empirical risk and the stochastic sensitive measure. The essence of LGE2LM is a quadratic problem without iterative process. Similar to ELM, all the parameters of this new algorithm are obtained without tuning, which makes the efficiency of LGE2LM much higher than those of traditional neural networks. Several experiments conducted on both artificial and real-world datasets show that LGE2LM has much better generalization capability and stronger robustness in comparison with ELM. [ABSTRACT FROM AUTHOR]