Search

Multi-label learning has attracted more and more researchers’ attention. It deals with the problem where each instance is associated with multiple labels simultaneously. Some methods improve the performance by constructing label-specific features. Specifically, the LIFTACE method constructs label-specific features by clustering ensemble techniques, which ignores the importance of label vectors and does not explore label correlations when constructing the classification model. In this paper, we propose a multi-label learning method called LF-LELC, which considers the importance of label vectors and constructs the classification model by considering label correlations. Firstly, it performs clustering on the positive instances and negative instances respectively. The number of clusters is set by the information contained in the label vectors. After that, it employs clustering ensemble techniques that consider label correlations to make the clustering results more stable and effective. Then, it constructs label-specific features for each label. Finally, it builds the classification model by exploring label correlations. The label set for each test example is predicted by the classification model. Experiments show that LF-LELC can achieve better performance by considering the importance of label vectors and the correlations among labels.

Low rank matrix approximation (LRMA) has a wide range of applications in computer vision and has drawn much attention in recent years. A typical nuclear norm minimization (NNM) is often used to solve a LRMA, but this is likely to overshrink the rank components due to having the same threshold. To address this problem, we propose a flexible and precise model named multi-band weighted l p norm minimization (MBWPNM). We have reformulated it into nonconvex l p norm subproblems under certain weight conditions, and we solve these subproblems via a generalized soft-thresholding algorithm. We then adopt MBWPNM for image (grayscale, color and multispectral) denoising. The proposed MBWPNM not only guarantees a more accurate approximation with a Schatten p-norm in case of a change of noise levels, but it also considers prior knowledge, for which different rank components have varying importance. Extensive experiments on additive white Gaussian noise removal and realistic noise removal demonstrate that the proposed MBWPNM achieves better performance than several state-of-the-art algorithms.

Domain adaptation can effectively transfer knowledge between domains with different distributions. Most existing methods use distribution alignment to mitigate the domain shift. But they typically align the marginal and conditional distributions with equal weights. This neglects the relative importance of different distribution alignments. In this paper, we propose a double weighted domain adaptation (DWDA) method, which employs new distribution alignment weighting and sample reweighting strategies. Specifically, the distribution alignment weighting strategy explores the relative importance of marginal and conditional distribution alignments, based on the maximum mean discrepancy; the sample reweighting strategy weights the source and target samples separately based on k-means clustering. The two strategies reinforce each other in the iterative optimization procedure, thus improving the overall performance. In addition, our method also considers the geometry structure preservation. The closed-form solution of the objective function is presented, and the computational complexity and convergence analysis are given. Experimental results demonstrate that DWDA outperforms state-of-the-art domain adaptation methods on several public datasets, and it also has good performance on an imbalanced dataset. Besides, DWDA is robust to a wide range of parameters. Moreover, the convergence curves show that DWDA generally converges rapidly within five iterations. We also evaluate the components of DWDA and show that each component is necessary. Finally, we compare the computational time with the methods of the recent three years to demonstrate the efficiency of DWDA.

Constraint Programming (CP) is an efficient technique for solving combinatorial (optimization) problems. In modern constraint solver, a CP Model is defined over reversible variables that take values in domains and propagators which filter the domains of the variables. Constraint propagation scheme schedules the propagators. A reasonable constraint propagation algorithm can improve the efficiency of solving CP problems. In this paper, we propose two efficient parallel propagation schemes based on multi-thread technique for table constraint. First, we give the formal definition of the parallel consistency and prove that the parallel propagation scheme is equivalent to the classic serial propagation scheme. Then, we propose two parallel propagation schemes: static submission and dynamic submission, which exploit work stealing thread pool and atomic operations to parallelize the classic propagation of table constraint. Finally, extensive experiments on various types of problems show that the two parallel schemes outperform their original serial version on a large number of instances. The results demonstrate the competitiveness of parallel propagation algorithms on solving extensional constraints.

Feature selection is an important method to reduce the number of attributes of high-dimensional data and an essential preprocess work in classification. It eliminates irrelevant, redundant, and noisy features improves the performance of the model and reduces the computational burden. Fruit fly optimization algorithm is a new algorithm proposed in recent years, which imitates the foraging behavior of fruit fly. To the best of our knowledge, it has not been systematically applied to feature selection. This paper uses the fruit fly optimization algorithm as a search strategy and designs a wrapper-based feature selection method, named binary improved fruit fly optimization algorithm (BIFFOA). Besides, four different strategies based on evolutionary population dynamics (EPD) and new mutation operators are employed to enhance the BIFFOA. The extensive experiments on 25 datasets (see Table 1) show that the performance of the BIFFOA is better than several state-of-the-art algorithms.

Object detection usually assumes that training and test data come from the same distribution, but the assumption is not always hold in practice. Due to domain shift problem, applying a trained detector to a new domain will lead to a great decrease in detection accuracy. Domain adaptive object detection has been adopted to maintain high detection accuracy in the face of various domain shift problems. Domain adaptive object detection methods mainly include adversarial-based methods, discrepancy-based methods, reconstruction-based methods, hybrid methods and others. Domain adaptive Faster RCNN is a classical adversarial-based method. In order to further improve the accuracy of domain adaptive object detection, we propose a method based on the Domain adaptive Faster RCNN called adaptive threshold cascade Faster RCNN (ATCFR). The ATCFR introduces the cascade strategy and adaptive threshold strategy. The cascade strategy improves the quality of bounding boxes and solves the problem of overfitting and mismatch in Faster RCNN. The adaptive threshold strategy ensures the balance of positive and negative samples and we don’t have to manually set the threshold as we did in cascade RCNN. In the end, we evaluate our new approach by using four classic datasets, including Cityscapes, Foggy Cityscapes, SIM 10k and KITTI. Experimental results show that our method has higher accuracy in variousdomain shift problems, compared with the state-of-the-art methods.

Market basket prediction, which is the basis of product recommendation systems, is the concept of predicting what customers will buy in the next shopping basket based on analysis of their historical shopping records. Although product recommendation systems develop rapidly and have good performance in practice, state-of-the-art algorithms still have plenty of room for improvement. In this paper, we propose a new algorithm combining pattern prediction and preference prediction. In pattern prediction, sequential rules, periodic patterns and association rules are mined and probability models are established based on their statistical characteristics, e.g., the distribution of periods of a periodic pattern, to make a more precise prediction. Products that have a higher probability will have priority to be recommended. If the quantity of recommended products is insufficient, then we make a preference prediction to select more products. Preference prediction is based on the frequency and tendency of products that appear in customers’ individual shopping records, where tendency is a new concept to reflect the evolution of customers’ shopping preferences. Experiments show that our algorithm outperforms those of the baseline methods and state-of-the-art methods on three of four real-world transaction sequence datasets.

Domain adaptation aims to mitigate the domain gap between the source and target domains so that knowledge can be transferred between domains. There are two key factors that determine the adaptation performance: transferability and discriminability. Transferability depends on the similarity of two domains. With transferability, the model learnt on the source domain can be used in the target domain. Discriminability indicates the separability of different classes. With discriminability, the adapted target features can be classified more accurately. Adversarial domain adaptation methods learn domain-invariant feature representations through adversarial learning. The domain-invariant feature representation guarantees the transferability. However, to obtain domain-invariant features, certain domain-specific information is suppressed, which may cause the loss of discriminability. To this end, we aim to enhance the discriminability by enriching the information contained in the domain-invariant features. We propose a Feature Concatenation for adversarial Domain Adaptation (FCDA) method. FCDA learns two feature extractors that can generate two different feature views for a sample. The concatenation of these two views is used as the feature representation of a sample, which we call the concatenation feature. Distribution alignment is performed on the concatenation features. We find that when the distributions of the concatenation features are aligned, the two feature views involved in a concatenation feature have different distributions. Thus, the concatenation feature contains more discriminative information, thereby enhancing the discriminative ability of the domain-invariant features. Experiments are carried out on four widely used datasets and FCDA exceeds some recent domain adaptation methods.

In machine learning and data mining tasks, feature selection is an important process of data preprocessing. Recent studies have shown that Binary Cuckoo Search Algorithm for Feature Selection (BCS [1]) has the better ability to classification and dimension reduction. However, by analyzing BCS algorithm, we notice that the randomness of initialization and the defects of fitness function severely weaken the classification performance and dimension reduction. Therefore, we propose a new feature selection algorithm FS\(\_\)CSO, which adopts the chaotic properties of the Chebyshev as a new initialization strategy to get the better original populations (solutions), and combines the information gain and the L1-norm as a new fitness function to accelerate the convergence of the algorithm. We validate FS\(\_\)CSO with various experimental data on small, medium and large datasets on the UCI dataset. In the experiment, FS\(\_\)CSO uses the KNN, J48 and SVM classifiers to guide the learning process. The experimental results show that FS\(\_\)CSO has a significant improvement in classification performance and dimension reduction. Comparing the FS\(\_\)CSO algorithm with the more efficient feature selection algorithms proposed in recent years, FS\(\_\)CSO is highly competitive in terms of accuracy and dimension reduction.

The coarse-grained arc consistency algorithms maintain a list of elements that need to be revised. It is well known that the performance of such algorithms is affected by the way in which the revisions are performed. In this paper, we propose a new variable-oriented propagation scheme that splits the propagation procedure into two independent phases. We show how it reduces the number of revisions and list operations. In experiments, we apply different revision ordering heuristics to this new variable-oriented scheme and compare them with the most efficient existing propagation schemes. The results from various structured and random problems show that the proposed propagation scheme reduces the number of revisions and offers speed-ups.

In this paper, we improved two classical degree-based variable ordering heuristics, $$\frac{\textit{Dom}}{\textit{Ddeg}}$$DomDdeg and $$\frac{\textit{Dom}}{\textit{Wdeg}}$$DomWdeg. We propose a method using the summation of constraint tightness in degree-based heuristics. We also propose two methods to calculate dynamic constraint tightness for binary extensional constraints and non-binary intensional constraints respectively. Our work shows how constraint tightness can be practically used to guide search. We performed a number of experiments on some benchmark instances. The results have shown that, the new heuristics improve the classical ones by both computational time and search tree nodes and they are more efficient than some other successful heuristics on the instances where the classical heuristics work well.

Model-based diagnosis in discrete event systems (DESs) is a major research topic in failure diagnosis, where diagnosability plays an important role in the construction of the diagnosis engine. To improve the solution efficiency for diagnosability, this paper proposes novel techniques to solve the problems of testing and optimizing for diagnosability. We propose a new concept, reverse twin plant, which is generated backwards from the final states of the DESs so there is no need to generate a complete copy of the DES model to determine the diagnosability. Such a design makes our testing algorithm much faster than existing methods. An efficient optimizing algorithm, which makes a non-diagnosable system diagnosable, is also proposed in the paper by expanding the minimal observable space with operation on just a part of the DES model. Examples and theoretical studies demonstrate the performance of the proposed designs.

Constraint satisfaction problem has many applications in Artificial Intelligence. Its interactive applications usually require advice from a system to help a user solve the problem. Based on maximal relaxations, the CorrectiveExp algorithm is a representative method to compute explanations. However, we found that the CorrectiveRelax algorithm, used by the CorrectiveExp algorithm to compute maximal relaxations, has a defect that it executes more consistency checks than necessary. It is very important to avoid these unnecessary consistency checks because in general each consistency check needs to resort to backtrack search. To tackle this problem, this paper proposes two improved algorithms to compute maximal relaxations, called CorrectiveRelaxReduced and CorrectiveRelaxDC respectively. The former utilizes the existing results of consistency checks to shrink the search scope for some inconsistent user constraints. Furthermore, we have proved that the number of consistency checks executed by the CorrectiveRelaxReduced algorithm is always less than or equals to that of the CorrectiveRelax algorithm. The latter uses a divide-and-conquer approach to avoid unnecessary consistency checks. Our experimental results show that the two improved algorithms execute less consistency checks than CorrectiveExp while computing maximal relaxations.

AC5TC is a generic value-based domain-consistency framework for GAC propagation algorithms, which stores the information of removed values with a propagation queue. Three efficient implementations of AC5TC algorithm have been proposed in [1]. One of these algorithms (called AC5TC-Tr) has an optimal time complexity theoretically, but its space complexity is inefficient because of the dynamic data structure. This paper proposes a novel algorithm based on AC5TC framework, called AC5TC-Alter, which leverages a more efficient search strategy to establish GAC. AC5TC-Alter accesses both allowed table and valid table alternately during the process of supports seeking, also our method is scalable since it does not need data structure to maintain the validity of tuples. The experimental results show that AC5TC-Alter outperforms most baseline algorithms on time complexity and AC5TC-Tr on space complexity.