This paper addresses classification tasks on a particular target domain in which labeled training data are only available from source domains different from (but related to) the target. Two closely ...related frameworks, domain adaptation and domain generalization, are concerned with such tasks, where the only difference between those frameworks is the availability of the unlabeled target data: domain adaptation can leverage unlabeled target information, while domain generalization cannot. We propose Scatter Component Analyis (SCA), a fast representation learning algorithm that can be applied to both domain adaptation and domain generalization. SCA is based on a simple geometrical measure, i.e., scatter, which operates on reproducing kernel Hilbert space. SCA finds a representation that trades between maximizing the separability of classes, minimizing the mismatch between domains, and maximizing the separability of data; each of which is quantified through scatter. The optimization problem of SCA can be reduced to a generalized eigenvalue problem, which results in a fast and exact solution. Comprehensive experiments on benchmark cross-domain object recognition datasets verify that SCA performs much faster than several state-of-the-art algorithms and also provides state-of-the-art classification accuracy in both domain adaptation and domain generalization. We also show that scatter can be used to establish a theoretical generalization bound in the case of domain adaptation.
Many evolutionary computation (EC) methods have been used to solve feature selection problems and they perform well on most small-scale feature selection problems. However, as the dimensionality of ...feature selection problems increases, the solution space increases exponentially. Meanwhile, there are more irrelevant features than relevant features in datasets, which leads to many local optima in the huge solution space. Therefore, the existing EC methods still suffer from the problem of stagnation in local optima on large-scale feature selection problems. Furthermore, large-scale feature selection problems with different datasets may have different properties. Thus, it may be of low performance to solve different large-scale feature selection problems with an existing EC method that has only one candidate solution generation strategy (CSGS). In addition, it is time-consuming to find a suitable EC method and corresponding suitable parameter values for a given large-scale feature selection problem if we want to solve it effectively and efficiently. In this article, we propose a self-adaptive particle swarm optimization (SaPSO) algorithm for feature selection, particularly for large-scale feature selection. First, an encoding scheme for the feature selection problem is employed in the SaPSO. Second, three important issues related to self-adaptive algorithms are investigated. After that, the SaPSO algorithm with a typical self-adaptive mechanism is proposed. The experimental results on 12 datasets show that the solution size obtained by the SaPSO algorithm is smaller than its EC counterparts on all datasets. The SaPSO algorithm performs better than its non-EC and EC counterparts in terms of classification accuracy not only on most training sets but also on most test sets. Furthermore, as the dimensionality of the feature selection problem increases, the advantages of SaPSO become more prominent. This highlights that the SaPSO algorithm is suitable for solving feature selection problems, particularly large-scale feature selection problems.
Using the method of blow-up analysis, we derive two Trudinger-Moser inequalities on a smooth line bundle over a compact Riemann surface with smooth boundary, as well as the existence of the ...corresponding extremals. Based on the earlier work of Li et al. (Calc Var Partial Differ Equ 28:59–83, 2007), the present paper generalizes Yang’s results (Pacific J Math 227:177–200, 2006 and Internat J Math 17:313–330, 2006). Moreover, we give the existence and uniform estimates of the Green section, which is very important in the blow-up analysis. Although much progress has been made with respect to the existence of extremals for Trudinger-Moser inequalities on Euclidean spaces and Riemann manifolds, there are only a little work on line bundle.
With a global search mechanism, particle swarm optimization (PSO) has shown promise in feature selection (FS). However, most of the current PSO-based FS methods use a fix-length representation, which ...is inflexible and limits the performance of PSO for FS. When applying these methods to high-dimensional data, it not only consumes a significant amount of memory but also requires a high computational cost. Overcoming this limitation enables PSO to work on data with much higher dimensionality which has become more and more popular with the advance of data collection technologies. In this paper, we propose the first variable-length PSO representation for FS, enabling particles to have different and shorter lengths, which defines smaller search space and therefore, improves the performance of PSO. By rearranging features in a descending order of their relevance, we facilitate particles with shorter lengths to achieve better classification performance. Furthermore, using the proposed length changing mechanism, PSO can jump out of local optima, further narrow the search space and focus its search on smaller and more fruitful area. These strategies enable PSO to reach better solutions in a shorter time. Results on ten high-dimensional datasets with varying difficulties show that the proposed variable-length PSO can achieve much smaller feature subsets with significantly higher classification performance in much shorter time than the fixed-length PSO methods. The proposed method also outperformed the compared non-PSO FS methods in most cases.
It is well known that the Adimurthi-Druet inequality admits extremal function, when the perturbation parameter is sufficiently small. As for the question of when extremal function does not exist, ...Mancini-Thizy first solved this problem by the method of energy estimate in (J. Differential Equations). After that Yang extended the work to a closed Riemann surface in (Sci. China Math.). In this paper, we generalize Yang’s result to a version involving
L
p
-norms for any
p
>
1
. Moreover, this work complements our result in (Acta Math. Sin.) and extends Wang’s result (Commun. Pure Appl. Anal.) in Euclidean space.
Feature selection is an essential step in various tasks, where filter feature selection algorithms are increasingly attractive due to their simplicity and fast speed. A common filter is to use mutual ...information to estimate the relationships between each feature and the class labels (mutual relevancy), and between each pair of features (mutual redundancy). This strategy has gained popularity resulting a variety of criteria based on mutual information. Other well-known strategies are to order each feature based on the nearest neighbor distance as in ReliefF, and based on the between-class variance and the within-class variance as in Fisher Score. However, each strategy comes with its own advantages and disadvantages. This paper proposes a new filter criterion inspired by the concepts of mutual information, ReliefF and Fisher Score. Instead of using mutual redundancy, the proposed criterion tries to choose the highest ranked features determined by ReliefF and Fisher Score while providing the mutual relevance between features and the class labels. Based on the proposed criterion, two new differential evolution (DE) based filter approaches are developed. While the former uses the proposed criterion as a single objective problem in a weighted manner, the latter considers the proposed criterion in a multi-objective design. Moreover, a well known mutual information feature selection approach (MIFS) based on maximum-relevance and minimum-redundancy is also adopted in single-objective and multi-objective DE algorithms for feature selection. The results show that the proposed criterion outperforms MIFS in both single objective and multi-objective DE frameworks. The results also indicate that considering feature selection as a multi-objective problem can generally provide better performance in terms of the feature subset size and the classification accuracy.
One of the major problems in Big Data is a large number of features or dimensions, which causes the issue of “the curse of dimensionality” when applying machine learning, especially classification ...algorithms. Feature selection is an important technique which selects small and informative feature subsets to improve the learning performance. Feature selection is not an easy task due to its large and complex search space. Recently, swarm intelligence techniques have gained much attention from the feature selection community because of their simplicity and potential global search ability. However, there has been no comprehensive surveys on swarm intelligence for feature selection in classification which is the most widely investigated area in feature selection. Only a few short surveys is this area are still lack of in-depth discussions on the state-of-the-art methods, and the strengths and limitations of existing methods, particularly in terms of the representation and search mechanisms, which are two key components in adapting swarm intelligence to address feature selection problems. This paper presents a comprehensive survey on the state-of-the-art works applying swarm intelligence to achieve feature selection in classification, with a focus on the representation and search mechanisms. The expectation is to present an overview of different kinds of state-of-the-art approaches together with their advantages and disadvantages, encourage researchers to investigate more advanced methods, provide practitioners guidances for choosing the appropriate methods to be used in real-world scenarios, and discuss potential limitations and issues for future research.
•Provide an overview of feature selection algorithms for classification.•Provide a discussion about the representation and search mechanisms of the most state-of-the-art swarm intelligence algorithms for feature selection.•Discuss the current limitations and future directions.
It is known that the Adimurthi–Druet inequality admits extremal function, when the perturbation parameter
α
is small. However, the problem that bothered us is when extremal function of the ...Adimurthi–Druet inequality does not exist. Recently, Mancini–Thizy (J. Differential Equations) first solved this problem by the method of energy estimate. After that, Yang (Sci. China Math.) extended the result to a closed Riemann surface. In this paper, on a compact Riemann surface with boundary, we consider the nonexistence of extremal function for a Trudinger–Moser inequality with the Neumann boundary condition. Moreover, this result complements our work in (Math. Inequal. Appl.).
Feature selection is an important task in data mining and machine learning to reduce the dimensionality of the data and increase the performance of an algorithm, such as a classification algorithm. ...However, feature selection is a challenging task due mainly to the large search space. A variety of methods have been applied to solve feature selection problems, where evolutionary computation (EC) techniques have recently gained much attention and shown some success. However, there are no comprehensive guidelines on the strengths and weaknesses of alternative approaches. This leads to a disjointed and fragmented field with ultimately lost opportunities for improving performance and successful applications. This paper presents a comprehensive survey of the state-of-the-art work on EC for feature selection, which identifies the contributions of these different algorithms. In addition, current issues and challenges are also discussed to identify promising areas for future research.
Convolutional neural networks (CNNs) have gained remarkable success on many image classification tasks in recent years. However, the performance of CNNs highly relies upon their architectures. For ...the most state-of-the-art CNNs, their architectures are often manually designed with expertise in both CNNs and the investigated problems. Therefore, it is difficult for users, who have no extended expertise in CNNs, to design optimal CNN architectures for their own image classification problems of interest. In this article, we propose an automatic CNN architecture design method by using genetic algorithms, to effectively address the image classification tasks. The most merit of the proposed algorithm remains in its "automatic" characteristic that users do not need domain knowledge of CNNs when using the proposed algorithm, while they can still obtain a promising CNN architecture for the given images. The proposed algorithm is validated on widely used benchmark image classification datasets, compared to the state-of-the-art peer competitors covering eight manually designed CNNs, seven automatic + manually tuning, and five automatic CNN architecture design algorithms. The experimental results indicate the proposed algorithm outperforms the existing automatic CNN architecture design algorithms in terms of classification accuracy, parameter numbers, and consumed computational resources. The proposed algorithm also shows the very comparable classification accuracy to the best one from manually designed and automatic + manually tuning CNNs, while consuming fewer computational resources.