Knowledge distillation has been widely applied as an effective model compression technique across various visual tasks. Currently, distillation methods suitable for object detection are typically realized through feature imitation in horizontal bounding box scenarios. In this paper, we propose an effective mask-based distillation scheme, MaskCrossKD. First, random pixels in the student features are masked, and then the mask is distilled through pixel reconstruction. The mask restricts the feature space of the student model, encouraging it to generate complete features that are consistent with the teacher model. Second, we introduce a simple distillation temperature control strategy, which dynamically adjusts the temperature to control the difficulty level of tasks during the student model’s learning process. Additionally, to adapt to rotated object detection, we propose a Rotational Distillation Loss (OKDL). This method avoids the discontinuity of rotated bounding box boundaries by mapping rotated boxes to a Gaussian distribution and designing a loss function based on the Gaussian Wasserstein distance. Experiments conducted on several public datasets across different models show significant improvements in detection performance compared to the teacher model. On the DOTA dataset, MaskCrossKD improved the average precision of FCOS ResNet-50 from 70.70% to 75.47% under the 1X strategy, and that of Oriented R-CNN ResNet-50 from 73.40% to 77.82%.

In the biomedical field, optimization algorithms are widely used in datasets such as pneumonia and cancer, often combined with machine learning methods for feature selection of datasets. We propose the Information Interconnected Harris hawks optimization algorithm (IIHHO), which enhances the optimization performance from three aspects: information transmission mode, exploitation mode, and exploration mode. First, the hawk group is divided into three layers: the leading layer, the guiding layer, and the ordinary layer, and information is transmitted between the layers so that the next generation is distributed in a better position. Then, the exploitation and exploration methods of the algorithm are improved to balance the exploitation and exploration stages of the algorithm. In order to verify the effectiveness of the algorithm, a comparative experiment was conducted on the CEC2022 test function to compare it with other advanced optimization algorithms. Subsequently, a feature selection experiment was carried out using the UCI data set. The resultant feature subset demonstrated good comprehensive performance, underscoring the high efficiency of our algorithm.

As an indispensable part of smart city development, traffic prediction's fundamental challenge is effectively modeling complex spatio-temporal dependencies in traffic data. Although previous work has made great efforts to learn the temporal dynamics and spatial dependencies of traffic, the following challenges still exist. Firstly, time series has multi-scale char-acteristics, meaning that traffic series display different trends at various time scales and change over time. Secondly, the spatial relationships within a traffic network are not singular. The traffic condition of a node is influenced not only by nearby nodes but also potentially by distant nodes. To this end, we propose a novel framework, spatio-temporal graph convolutional networks considering multiple spatio-temporal information (STGCN-MI), for traffic prediction. Specifically, in the temporal dimension, we design a multi-scale local multi-head self-attention module. It more appropriately assigns correlation strengths to data pairs in the temporal dimension by significantly enhancing the ability to represent trends in the series at different time scales, thereby capturing dynamic temporal correlations more accurately. In the spatial dimension, we develop a fusion graph convolution module, which mines the multiple spatial dependencies in the traffic network and fuses them adaptively. Extensive experiments on two real-world datasets demonstrate that the effectiveness and superiority of our methods.

Hyperspectral image classification is an important but challenging task. Conventional convolutional neural networks (CNNs) are able to extract local spectral spatial features but ignore long-range dependencies and global features. To address this problem, we propose a new model combining 3D-CNN and Convolutional Vision Transformer, aiming to improve the performance of the image recognition task by utilizing the advantages of CNN in local feature extraction while retaining the advantages of Transformer in long-range dependency processing. Our model is tested on three publicly available hyperspectral image datasets, and the results show that our model outperforms other state-of-the-art models in terms of classification accuracy and robustness. The source code for our work is available at [https://github.com/Dreamvai/ViT-Convolution]. The model proposed in this letter provides a new idea for hyperspectral image classification and expands a new field for the application of convolutional transformers. In the future, we intend to further explore the performance of the convolutional transformer and the possibility of combining it with other types of data.

In this letter, we propose a robust cross-drone multi-target association algorithm that aims to associate the identities of targets detected by different drones with overlapping FOVs. Unlike other methods using image features (target appearance features and background features), our approach makes full use of the three-dimensional spatial consistency (TSC) between targets. It has wider applicability in deployed overlapping cross-drone systems with unreliable image features. Specifically, we first explore target correlation from the perspective of three-dimensional space, and propose an association method that integrates position and altitude consistency constraints, without relying on image features. After that, we mine topological mapping relationships (TMR) between cameras from the preliminary results and propose a TMR-based re-association approach using cost evaluation to optimize the preliminary results, which effectively improve association accuracy. Extensive experimental evaluations on Airsim-based dataset verify the effectiveness and robustness of our proposed method.

With the widespread use of GPS-enabled smart devices and the increased availability of wireless networks, mobile crowdsourcing system (MCS) has recently been proposed as a framework that automatically requests workers to perform location sensitive tasks. In the task allocation problem of MCS, existing algorithms lack consideration of the impact of nonadjacent and discontinuous execution of tasks on task allocation. Workers may have different task preferences in different time periods. Nonadjacent and discontinuous tasks provide important correlations for understanding workers’ behavior. This article introduces a novel task allocation algorithm based on spatiotemporal attention network (STATA). STATA takes into account factors such as the spatiotemporal distribution of tasks and workers as well as the location preferences and abilities of workers, and integrates them into a unified network for modeling. First, all historical tasks performed by workers are aggregated to obtain the correlation of all historical tasks. Then, the most plausible candidate tasks are recalled from the weighted representation for allocation. STATA utilizes the spatiotemporal attention mechanism to capture the relationship between these factors, ultimately improving the accuracy of task allocation. Extensive experiments demonstrate that the STATA model exhibits superior performance in terms of task allocation accuracy and practical application capabilities.

Traditional training models often experience significant performance drops on test sets when the training and testing data distributions differ. To address this domain shift problem, we propose D3Net, a domain generalization (DG) classification network for hyperspectral images (HSIs) based on generative adversarial networks (GANs). Specifically, D3Net consists mainly of a generator and a discriminator. The generator extracts domain-invariant information from the source domain to generate data with core classification features, while the discriminator employs a dual-confidence model to enhance the capture of domain-invariant features. Through adversarial iterations, the model is able to adapt to the domain shift effects of unknown data. Unlike existing DG methods that rely on random perturbations for data augmentation, D3Net utilizes learnable convolutional neural networks (CNNs) rather than randomization to enhance the model’s learning capability. We conducted cross-scene classification experiments on datasets from Houston, Pavia, and Indiana, and the results demonstrate the effectiveness of our approach. The code for D3Net is available at:https://github.com/gmsjzyq123/D3Net.

In hyperspectral image (HSI) classification tasks, extracting surface features is crucial, but the complexity of HSI brings considerable challenges. Traditional convolutional neural networks (CNNs) and vision transformer (ViT) perform well but have large parameters and are time-consuming. In contrast, graph convolutional neural networks (GCNs) are fast to train with few parameters. The simple linear iterative clustering (SLIC) segmentation algorithm generates pixel-level graph data and superpixel-level graph data. Single-branch GCNs learn only one type of graph data resulting in insufficient feature richness. In two-branch GCNs, one branch extracts local spatial-spectral features from pixel-level data, another branch extracts global spatial-spectral features from superpixel-level data, and finally, the two branches are combined. However, the existing two-branch GCNs do not focus on the close fusion of local and global features. Therefore, in this article, we propose new two-branch GCNs with local and global feature fusion (LG-GCNs). The LG-GCN model proposed in this study not only has a short training time but also has a high accuracy. The LG-GCN can capture more comprehensively the local and global spatial-spectral semantic information of HSI. The experimental results on three publicly available HSI datasets show that LG-GCN outperforms other state-of-the-art models in terms of classification accuracy and robustness. In addition, it has a shorter training time and is more resistant to noise. The source code of the LG-GCN model is available athttps://github.com/Dreamvai/LG-GCN.

As one of the swarm intelligence algorithms, Salp Swarm Algorithm (SSA) is widely used in feature selection problems due to its fast convergence speed and simple structure. However, SSA still has the disadvantages of poor population diversity and being prone to fall into local optimum. To solve these problems, this study proposes a cooperative co-evolutionary salp swarm feature selection algorithm (CESSA) for classification. Firstly, different populations are divided according to the characteristics of the salps, which solves the drawback of a single chain in the SSA algorithm. Multiple populations are generated from mutual information (MI)-based, random, chaotic map-based initialization corresponding to different search tasks. Learning the habitats of the solitary salps and modeling them ensures the exploration ability of the algorithm. Finally, a cooperative strategy is presented to integrate the advantages of different populations to achieve cooperative co-evolution. CESSA is evaluated on fourteen datasets and compared with six feature selection methods. The results show that CESSA can select smaller feature subsets with higher classification accuracy in most cases.

Mayfly algorithm (MA) is a new type of intelligent optimization algorithm with excellent search ability and a broad application prospect in feature selection. However, as the dimension of the data increases, MA also has the problem of weak search ability and falling into the local optimal solution. To solve these problems, this paper combines the idea of collaborative evolution and proposes an improved Mayfly algorithm (CEIMA) suitable for feature selection problems. The CEIMA algorithm first divides the population into two sub-populations and initializes them in different ways to increase the diversity of the population. During evolution, information sharing is achieved through an information transmission mechanism between sub-populations, achieving collaborative evolution between sub-populations to enhance the global search ability of CEIMA. Finally, the mRMR-m strategy is proposed to mutate the global optimal individual based on feature importance, improving the ability of CEIMA to escape from local optimal solutions. Through a comparison with 7 feature selection algorithms on 12 UCI datasets, the effectiveness of CEIMA is proven.

Task pricing is an important step for crowd-sourcing platforms to solve profit-driven task allocation and maximize profits. Most of the existing researches only carry out algorithm design on the premise of fully determining the scene information. However, due to the interference of many factors in the real scene, information such as workers and task costs in the scene is usually not completely uncertain. To solve the above problems, a spatial crowdsourcing task pricing algorithm is proposed. Firstly, the algorithm uses the proposed dimension-up gray wolf algorithm and support vector regression(DU_GWO-SVR) to predict the task price, and then sets the price based on the obtained price. In order to solve the instability of average price and matching number caused by dynamic supply and demand, an adjustment mechanism is designed to stabilize the average price of tasks. The experiment uses a real data set-the New York taxi data set, and compares it with the classic greedy algorithm and binary matching algorithm. The experimental results show that the matching rate of the proposed algorithm is above 75% when the scene information is not completely uncertain.

Ship detection in large-scale synthetic aperture radar (SAR) images has essential value in both military and civilian applications. However, due to the complexity of the background and the simplicity of the texture, ship detection in large-scale SAR images is prone to false alarms, such as similar-shaped reefs, islands, sea clutter, and inland buildings. This article proposes a multisize inference fusion framework to eliminate false alarms and improve the overall performance of ship detection in large-scale SAR images. In this framework, a multisize slicer is proposed to expand the scale range of image expression. Then, a detection model library is built to keep various types of models for different task scenarios and requirements. Finally, two subapproaches are proposed for false alarm elimination, namely, pixel feature filtering (FAE-pff) and multisource fusion (FAE-msf), to reduce false detection results in the output of the detection model. FAE-pff calculates how obvious each target is relative to the background and eliminates less obvious results. FAE-msf obtains bounding boxes and corresponding confidences from multiple inference sources and fuses them through weighting and updating them to achieve complementation and enhancement of information. Various experiments were conducted to evaluate the performance of each module qualitatively and quantitatively. It proves the effectiveness of the proposed framework, which can achieve more correct detections while greatly reducing erroneous detections.

Momentum term technique is an efficient resolution to accelerate the convergence speed of adaptive blind source separation (BSS) algorithms, however, the BSS algorithm combined with a momentum term also suffers from the tradeoff between the fast convergence speed and small misadjustment error. In order to alleviate such compromise, an optimal variable momentum factor method is used to boost the separating performance of the nonlinear principal component analysis (NPCA) BSS algorithm. At first, by using the projection approximation, the cost function of the NPCA algorithm can be represented as a quadratic function of the momentum factor. Then the optimal momentum factor is obtained on the basis of the gradient decent technique, which makes the cost function descend in the fastest way during each iteration. Simulation experiment results manifest that the modified algorithm can improve the convergence speed and decrease the final misadjustment error more effective compared with the classical NPCA algorithms and the fixed momentum factor NPCA algorithms.

Earthquake prediction based on extreme imbalanced precursor data is a challenging task for standard algorithms. Since even if an area is in an earthquake-prone zone, the proportion of days with earthquakes per year is still a minority. The general method is to generate more artificial data for the minority class that is the earthquake occurrence data. But the most popular oversampling methods generate synthetic samples along line segments that join minority class instances, which is not suitable for earthquake precursor data. In this paper, we propose a Safe Zone Synthetic Minority Oversampling Technique (SZ-SMOTE) oversampling method as an enhancement of the SMOTE data generation mechanism. SZ-SMOTE generates synthetic samples with a concentration mechanism in the hyper-sphere area around each selected minority instances. The performance of SZ-SMOTE is compared against no oversampling, SMOTE and its popular modifications adaptive synthetic sampling (ADASYN) and borderline SMOTE (B-SMOTE) on six different classifiers. The experiment results show that the quality of earthquake prediction using SZ-SMOTE as oversampling algorithm significantly outperforms that of using the other oversampling algorithms.

To stimulate large‐scale users to participate in the big data construction of IoT (internet of things), auction mechanisms based on game theory are used to select participants and calculate the corresponding reward in the process of crowdsensing data collection from IoT. In online auctions, bidders bid many times and increase their bid price. All the bidders want to maximize their utility in auctions. An effective incentive mechanism can maximize social welfare in online auctions. It is complicated for auction platforms to calculate social welfare and the utility of each bidder’s bidding items in online auctions. In this paper, a transaction trade‐off utility incentive mechanism is introduced. Based on the transaction trade‐off utility incentive mechanism, it can make the forecasting process consistent with bidding behaviors. Furthermore, an end‐price dynamic forecasting agent is proposed for predicting end prices of online auctions. The agent develops a novel trade‐off methodology for classifying online auctions by using the transaction trade‐off utility function to measure the distance of auction items in KNN. Then, it predicts the end prices of online auctions by regression. The experimental results demonstrate that an online auction process considering the transaction utility is more consistent with the behaviors of bidders, and the proposed prediction algorithm can obtain higher prediction accuracy.

Particle swarm optimization (PSO) has a bright future in feature selection (FS). However, with the increase of data set dimension, the search space becomes larger, PSO is easy to fall into local optima and brings a lot of time and space overhead. It is still a big challenge to apply PSO to thousands of feature data sets. In this paper, the features are reordered to ensure that the shorter particles can get better classification performance. We propose a subset length constraint mechanism to reduce the number of selected features of particles gradually, so as to introduce particles into smaller and more effective space and jump out of local optima. Although the short running time of filter FS is more suitable for high-dimensional feature selection, the classification accuracy of filter FS is generally lower than that of wrapper FS. In order to achieve better classification performance in a short time, we propose a dual global optimal (Gbest) updating model. PSO has been modified. Compared with six algorithms on six high-dimensional data sets. The results show that the new dual Gbest update mechanism based on length restriction mechanism has higher efficiency and accuracy.

Unsupervised domain adaption aims to reduce the divergence between the source domain and the target domain. The final objective is to learn domain‐invariant features from both domains that get the minimised expected error on the target domain. The divergence between domains which is also called domain shift is mainly between the distributions of domains' samples. Additionally, the label shift is also a tricky challenge in domain adaptation. In this study, domain‐invariant adversarial learning with conditional distribution alignment is proposed to alleviate the effect of domain shift with label shift. To obtain the domain‐invariant features, the proposed method modifies adversarial auto‐encoder architecture and performs semi‐supervised learning to enlarge the inter‐class discrepancy. The marginal distribution is aligned in the adversarial learning process of extracting domain‐invariant features. Meanwhile, the label information is incorporated in this way to align the conditional distribution. The proposed work also theoretically analyses the generalisation bound of the proposed model. Finally, the proposed method is evaluated based on several domain adaptation tasks, including digit classification and object recognition, and achieves state‐of‐the‐art performance.

This paper proposes a feature selection method based on adaptive differential evolution-ISHADEFS. By using mutual information and Pearson correlation coefficient to construct a new fitness function, the enhancement effect of correlation and similarity is used to improve the efficiency and accuracy of filtering. A triangular mutation operator is improved, and the operator diversity is enhanced by the operator adaptive structure to improve the performance of the algorithm. The experimental results show that compared to conventional filtering methods and original evolution methods, ISHADEFS can achieve better classification performance.

With the development of social life, prediction becomes more and more important. As an emerging sequence modeling model, the temporal convolutional network has been proven to outperform on tasks such as audio synthesis and natural language processing. But it is rarely used for time series prediction. In this paper, we apply the temporal convolutional network into the time series prediction problem. Gated linear units allow the gradient to propagate through the linear unit without scaling so we introduce it in temporal convolutional networks. In order to extract more useful features, we propose a multi-channel gated temporal convolution network model. We use the model for stock closing price prediction, Mackey-Glass time series data prediction, PM2.5 prediction, and appliances energy prediction. The experimental results show that compared with the traditional methods, LSTM, and GRU, the temporal convolutional network, gated temporal convolutional network and multi-channel gated temporal convolution network converge faster and have better performance.

Image denoising is an important pre-processing step in image processing. Different algorithms have been proposed in past few decades with varying denoising performances. In recent years, the image-based sparse representation can better characterize the essential features of images. Image sparse representation based on adaptive overcomplete dictionary can satisfy the sparsity and image noise separability, and has been successfully applied to image denoising, forming a kind of image denoising algorithms based on sparse representation. K-singular Value Decomposition algorithm is currently the most representative, the most widely applied adaptive learning dictionary image denoising algorithm. Based on the K-Singular Value Decomposition algorithm, this paper proposes an image denoising algorithm based on dictionary learning of mean correction atoms. In the dictionary initialization process, the dictionary learning scheme of mean corrected atoms is proposed to effectively suppress noisy atoms. The experimental results show that this algorithm can obtain better image restoration quality than these similar algorithms.

Video-based behavior recognition is a challenging research topic. The three dimensional convolution neural network (3D CNN) is effectively adopted to capture features from videos directly. 3D CNN is extended by two-dimensional convolution neural network, in which a time dimension is added. 3D CNN is better than two-dimensional convolution network in expressing effective motion information, and it has certain advantages. In order to make better use of the valuable features extracted from the original video information, only stacked RGB frame data sets can be used as the input of network. Ensembling 3D CNN framework for video recognition is proposed in the paper. Firstly, the pre-training model of Sports-1M is initialized firstly, and a 3D convolution neural network based on multi-level feature fusion is constructed. . The final high-dimensional feature combination is obtained by fusing multiple convolution features. Then 3D convolutional neural network based on ensemble learning is proposed to increase motion information, enrich motion features and enhance the robustness of single feature representation. Three incomplete training data sets are obtained by Bagging algorithm. To get different networks, three data sets are employed to train three 3D convolution neural networks respectively, and the output of the three networks is integrated. The output features of the three networks are input into the SVM classifier through the Stacking algorithm and the final results are obtained. The integration effects of different ensemble methods are compared. The experimental results show that the method of this work can improve recognition accuracy on UCF-101 data set effectively.

Recently deep learning has revolutionized the field of machine learning, for pattern recognition in particular. A deep neural network (DNN) requires a large number of labeled training samples, and the recognition accuracy is truly impressive, sometimes outperforming humans. Neurons in an artificial neural network (ANN) are modeled by a single, static, continuous-valued activation function. However, biological neurons employ discrete spikes to compute and transmit information. The information can be encoded by the spike times, spike rates, and spike phase, etc. As the third generation artificial neural networks, spiking neural networks (SNNs) are more closely mimic natural neural networks. SNNs can achieve the same goals as ANNs, and it has the ability to build a large-scale network structure (i.e. deep spiking neural network) to accomplish complex tasks. In this paper, a state-of-the-art manner, differential evolving spiking neural network (DESNN), is proposed for pattern classification. The XOR task, Iris data, and hand-written digits classification task on MNIST are used to validate the proposed training method. The experimental results show that the algorithm used in this work applies the fewer neurons and it is effective for pattern classification tasks.

In this paper, an effective differential evolution is proposed with binary strategy to solve feature selection problem. Firstly, a new binary mutation operator and a binary crossover operator are designed. The two-stage adaptive strategy is constructed in the mutation operator to generate new individuals to improve the diversity of the population. Then, the adaptive cross-parameter selection based on individual is developed in the crossover operator to fully exploit each potential optimal individual. Finally, six benchmark data sets are adopted to evaluate the effectiveness of the proposed algorithm. The experimental results show that the proposed algorithm significantly improves the classification accuracy, reduction rate and time cost.

Crowdsourcing is the perfect show of collective intelligence, and the key of finishing perfectly the crowdsourcing task is to allocate the appropriate task to the appropriate worker. Now the most of crowdsourcing platforms select tasks through tasks search, but it is short of individual recommendation of tasks. Tag-semantic task recommendation model based on deep learning is proposed in the paper. In this paper, the similarity of word vectors is computed, and the semantic tags similar matrix database is established based on the Word2vec deep learning. The task recommending model is established based on semantic tags to achieve the individual recommendation of crowdsourcing tasks. Through computing the similarity of tags, the relevance between task and worker is obtained, which improves the robustness of task recommendation. Through conducting comparison experiments on Tianpeng web dataset, the effectiveness and applicability of the proposed model are verified.

To solve the precocious and convergence problem, a reverse-learning particle swarm optimization algorithm based on niching technology (NRPSO) was proposed. The algorithm not only retains the historically optimal position of each particle, but also retains the historically worst position of the particle. When the particle has been trapped into a local optimum, the reverse-learning mechanism is adopted. Niche is generated through fuzzy clustering. The simulated annealing method is used inside the niches to guide excellent particles for local mining and learning. The reverse-learning mechanism is adopted between the niches and the particles are guided by the niche territory with low average of fitness to jump out of the local optimum. The experiment results on a set of benchmark functions with different dimensions show that the optimization performance, search efficiency and convergence speed of NRPSO algorithm are much better than other modified PSO algorithm.

With the rapid growth of information, feature selection has become an important step before data classification. Evolutionary algorithms can be used to solve feature selection given their effective search capabilities since the traditional feature selection cann't consider the combination effect. In this paper, we present a novel filter feature selection using a niching binary particle swarm optimization. The entire population is divided into several niche groups to maintain the diversity of evolutionary environment. A new framework based on three different kinds of topologies is proposed which can avoid falling into the local optimum and enhance global search capabilities. In this framework, when the optimal fitness value has stagnated for 20 generations, the connection between particles in each niche group and the connection between niche center particles will change to improve the optimal fitness value. The above procedure can produce a subset of features. In order to verify the effectiveness of the proposed algorithm, we have tested on five data sets in the UCI database. The experimental results show that the proposed algorithm can effectively search for the feature space and verify the efficiency of the obtained feature subsets under different classifiers.

How does a search engine company decide which advertisements to display for each query to maximize its revenue? This turns out to be a generalization of the online bipartite matching problem. In this paper, search engines decide the strategy to allocate resources with an advertiser credibility factor under budget constraints. Based on the optimal algorithm, which is the notion of a trade-off revealing LP, this paper remains the competitive ratio as 1 - 1/e with an advertiser credibility factor. During the ranking in the keywords auctions, CTR(Click Through Rate) and credibility factor are added to the trade-off function. In the long term, users tend to use the search engine with high credibility, which will bring greater revenues. The search engine will attract advertisers to bid on keywords and improve their credibility actively.

In order to improve the accuracy of support vector machine (SVM) classification of wetland remote sensing images, the selection of kernel function parameters in support vector machines becomes an effective approach. In this paper, Particle Swarm Optimization and Genetic Algorithms (PSO-GA) coevolutionary algorithm are used to optimize the SVM parameters.Because of the complementarity of evolutionary features between PSO and GA, this algorithm is combined with PSO and GA to improve the convergence speed and realize the optimization of depth and breadth. Experimental results show that SVM with PSO-GA co-evolutionary algorithm can achieve high classification accuracy in finite iteration times compared with existing intelligent optimization algorithms.

The dendritic cells algorithm (DCA) is an algorithm which simulates antigen presentation of dendritic cells in biological sciences. As a binary classifier, the DCA can classify input data items as normal and abnormal one quickly and efficiently. DCA uses the Principal Component Analysis (PCA) to do feature extraction and signal categorization. However, using PCA presents a limitation as the signals extraction are depend on artificial determination. To overcome this limitation, this study proposes a numerical differentiation based dendritic cell model. The proposed model introduces numerical differentiation theory which can describe that data changes will lead to danger, and extracts signals adaptively, through deep data analysis with respect to change and adaptive. Indeed, DCA is sensitive to the input data sequence for each DC gathers multiple antigens over a period of time. Therefore, the study proposes a numerical differentiation based DCA (NDDCA). DC samples antigen randomly and dynamically to overcome the sensitivity to the order of input data, which makes the classification of NDDCA clearer and only focus on the classification data source. Experiments on real data sets show that our new approach which focuses on unordered data binary classification problems delivers more accurate results.

Quantum-behaved particle swarm optimization (QPSO) is a novel variant of particle swarm optimization (PSO), inspired by quantum mechanics. Compared with traditional PSO, the QPSO algorithm guarantees global convergence and has less number of controlling parameters. However, QPSO is likely to get trapped into a local optimum because of using a single search strategy. This paper proposes a cooperative quantum particle swarm optimization (CGQPSO) algorithm based on multiple groups which apply different search strategies. The diversity of search strategies balances exploration and exploitation and avoids the local optimal problem. A cooperative mechanism, such as competition and cooperation, is introduced to implement the adaptive adjustment of a particle swarm. The dynamic adaptability of the particle swarm can adjust different search strategies according to a specific problem. The experimental results of 10 benchmark functions show that the proposed CGQPSO outperforms than other QPSO variants in terms of the performance and robustness.

Biogeography-based optimization (BBO) is a recently-developed algorithm that uses migration to share information among candidate solutions. We use differential evolution algorithm's mutation operator to improve the individual migration operator, and take an adaptive method in setting the value of the scaling factor. The new individual migration is combined with two traditional gene migrations, thus we get a new multiple migrations operator. The biogeography-based optimization with multiple migrations (HLBBO) is proposed based on this new operator. Experiments have been conducted on 25 benchmarks from the 2005 Congress on Evolutionary Computation. Compared with BBO algorithm and linearized BBO, the results show that the proposed algorithm HLBBO can improve the convergence speed and solution accuracy. And the boxplot of the best fitness value show the algorithm' s stability.

As a variant of the classical knapsack problem, the multi-dimension multi-choice knapsack problem (MMKP) is widely used in practical applications. It is a NP-complete problem, the exact solution of MMKP cannot be founded in polynomial-time. As one of the heuristic algorithms, quantum particle swarm optimization (QPSO) algorithm provides a sight to get the approximately optimal result for MMKP. However, due to the multiple constraints among dimensions and the dispersing feasible regions, QPSO tends to fall into local convergence. Hence a modified diversity reserved QPSO algorithm for MMKP is proposed in this paper: (i) to measure the availability of a particle by comparing the position between itself and the next alternative during the generation; (ii) import a position disturbance operator to increase the diversity of population. Experiments demonstrate that the proposed evolutionary algorithm could find better near-optimal results. And the analysis of convergence and execution time suggest that the probability of local convergence is declined in our algorithm.

Curling-match arrangement is a multi-constrained optimization problem in the real world. An improved partheno-genetic algorithm is used for solving the problem in this paper. In order to handle the complicated relationships among the particular constraints in curling-match, an eliminate-selection strategy is proposed to increase population diversity. Two genetic operators, targeted self-crossover operator and fixed-random self-crossover operator, are designed to ensure that the algorithm can convergence rapidly. With bi-level optimization, the improved partheno-genetic algorithm enhances its search ability. An orthogonal method is used to obtain the algorithm parameters. Simulation results demonstrate that the improved algorithm can solve the curling-match multi-constrained optimization problem efficiently.

Different mutation operators such as Gaussian, Cauchy and Lévy mutations have been proposed in evolutionary programming. According to the no free lunch theorem, operators are only efficient within certain fitness landscapes. Therefore the mixed strategy, integrating several mutation operators into a single algorithm, is a nature development in order to combine the advantages of different operators. Based on Shapley value, this paper presents a new mixed strategy evolutionary programming algorithm. It employs Gaussian, Cauchy and Lévy mutation operators and uses Shapley value to assign weights to these three operators. Then evolutionary programming using the new mixed strategy is tested on a set of 22 benchmark problems. The performance of the new mixed strategy is compared with other two mixed mutation strategies and three pure strategies. The experimental results show that the new mixed strategy has achieved an acceptable accuracy.

Feature selection is an important part of machine learning and data mining which may enhance the speed and the performance of learning and mining algorithms. Given certain criteria to evaluate features, the problem of feature selection can be regarded as an optimization problem. Therefore, evolutionary algorithms can be used to solve such a kind of optimization problems. In this paper, we present a novel feature subset selection approach based on the framework of genetic algorithms. Two new mutation operators are constructed using the standard deviation of candidate features and the cardinality of candidate feature subsets. Then, a filter feature subset selection approach using a dynamic mixed strategy is proposed, which combines the new mutation operators with the single-point mutation operator. The new approach can not only dynamically adjust the probability distribution over these three mutation operators, but also maintain the combined effects of feature subsets as a whole fitness evaluation. The proposed approach is able to quickly escape from local optimal feature subsets and to obtain smaller scale subsets than evolutionary algorithms using a single mutation operator. Experiments have been implemented on six standard UCI datasets and the proposed algorithm is compared with other classical algorithms. The comparison outcomes confirm the effectiveness of our approach.

To address computational complexity of winner determination in combinatorial auction, a new co-evolutionary algorithms is developed based on combining mixed mutation with self-organization optimization for finding high quality solutions quickly. Mixed mutation strategy can select adaptively mutation operators which are suitable for discrete space to maintain population diversity, self-organization optimization makes the search to jump out of local optima. This paper investigates two combination methods of mixed mutation and self-organization optimization, the results of experiment show the better performance of the second way (MMSEO2) that self-organization optimization is added to mixed mutation strategy set as a pure mutation operator. We compare the proposed algorithm with current well-known approximate algorithms for winner determination problem, and demonstrate that the proposed algorithm MMSEO2 produces competitive results and finds better solutions than other algorithms for large problem sizes.

In order to precisely study the advertisers' bidding behavior, in this paper, we proposed a HMM(Hidden Markov Model) forecasting model using the historical auction data of advertisers, and predicted the advertisers' bidding sequences in the future with this model. In the process of establishing HMM model for advertisers' bidding behavior, we define the bidding as the hidden variable, and define the position that advertiser obtained as the observable variable in this model. In order to verify the effectiveness of this approach, we compared this method with existing Bayesian forecasting model, and found that HMM model predicted advertisers' bid closer to the actual auction; In addition, we used this method in the TAC/AA(Trading Agent Competition/Ad Auctions) game platform, and finally achieved good results. Therefore, HMM bidding behavior model can well simulate advertisers' bidding sequence, and provide a very good forecasting method for advertisers' bidding, and also help search engines to develop appropriate auction mechanism by predicting advertisers' bidding sequences.

A pure strategy metaheuristic is one that applies the same search method at each generation of the algorithm. A mixed strategy metaheuristic is one that selects a search method probabilistically from a set of strategies at each generation. For example, a classical genetic algorithm, that applies mutation with probability 0.9 and crossover with probability 0.1, belong to mixed strategy heuristics. A (1+1) evolutionary algorithm using mutation but no crossover is a pure strategy metaheuristic. The purpose of this paper is to compare the performance between mixed strategy and pure strategy metaheuristics. The main results of the current paper are summarised as follows. (1) We construct two novel mixed strategy evolutionary algorithms for solving the 0-1 knapsack problem. Experimental results show that the mixed strategy algorithms may find better solutions than pure strategy algorithms in up to 77.8% instances through experiments. (2) We establish a sufficient and necessary condition when the expected runtime time of mixed strategy metaheuristics is smaller that that of pure strategy mixed strategy metaheuristics.

Adaptive equalization technology requires a long training sequence to update the parameters of the taps by gradient descent method step by step. In order to decrease the number of training sequence, this paper proposes an improved version of the classical differential evolution algorithm for adaptive equalizer to estimate the parameters, in which two trial vectors are created by crossover operator. The modified algorithm speeds up the convergence rate and improves the convergence precision through the evolution of multi-generation in the situation of a short training set. Compared with the traditional least mean squares (LMS) algorithm and the classical differential evolution (CDE) algorithm, the modified algorithm can switch to data transmission mode from the training mode much earlier; at the same time improve the efficiency of the transmission greatly. The simulation results have confirmed that the proposed algorithm achieves the faster convergence rate, the lower misadjustment and the less symbol error rate than the LMS algorithm and CDE algorithm in 4-PAM and 16-QAM signal systems.

Genetic algorithms (GAs) have long been used for large join query optimization (LJQO). Previous work takes all queries as based on one granularity to optimize GAs and compares their efficiency with other query optimization algorithms. However, we believe that large join queries are based on a granularity that is too large (1) to optimize GAs and (2) to compare the efficiency of different randomized optimization algorithms. Besides, while previous work only discusses the efficiency of basic GAs for LJQO, we believe that hybrid GAs reduce search space to improve GAs efficiency.

In recent years, kNN algorithm is paid attention by many researchers and is proved one of the best text categorization algorithms. Text categorization is according to training set, which is assigned class label to decide a new document, which is not assigned class label belongs to some kind of document. But for a classifier, text preprocessing is the bottleneck of categorization. In the original feature space, there are always thousands upon thousands words. The dimension of feature space is very high. So in this paper, we adopt a new feature weight method---- improved Gini index to reduce the dimension of feature space and improve the categorization precision. In addition, we discuss the improvement of decision rule and dimension selection. We design an adaptive fuzzy kNN text classifier. Here the adaptive indicate the adaptive of dimension selection. The experiment results show that our algorithm is effective and feasible.

With the development of WWW (World Wide Web), more and more people surf on the Web, more and more Web sites provide forum for people to publish their reviews, so there are many reviews on some special topic in many forums. Our system mainly aims at the reviews of customer for some product. The data set comes from forum data or emails or any other form of reviews. The mining result can guide the company's CEO to make science decision for company's products research and market development and the mining result can guide the customers to purchase more satisfying products. Our system mainly adopts data mining technology, natural language processing technology, Web text mining technology and so on. In the realization of our system, we adopt role based concepts and theory, this makes the realization more reasonable and more efficiency and this makes the system function more perfect and more science.

The computer defense immune system (GDIS) is an artificial immune system for detecting computer viruses and network intrusions. We present a simple chromosome-based evaluation model for GDIS. In this model, the genotype space is a linear number sequence, and a digital pattern sequence is produced as the phenotype space using a number of mechanisms, including pattern mining and genetic algorithms. We present a range of experiment analyses to show the higher efficiency and stronger immunity of this model, improving the rate of successful prediction in intrusion detection in GDIS. The detectors generated may have higher coverage, hence would impose lower communications and computational loads on systems in which they were incorporated.