Search

The algorithms on learning to rank can traditionally be categorized as three classes including point-wise, pair-wise and list-wise. In our work, we focus on the regression-based method for the multi-partite ranking problems due to the efficiency of the point-wise methods. We proposed two ranking algorithms with the real AdaBoost and the discrete AdaBoost, which compute the expectation of the ratings with the estimation of the pseudo posterior probabilities. We found that it can be explained in the framework of the regression with the squared loss. It is more easily implemented than the previous McRank method since the algorithm adopts the decision stump as the weak leaner instead of the regression tree. In the fifteen benchmark datasets, our methods achieve better performance than the pair-wise method RankBoost under the C-index, NDCG and variant of NDCG measures. It has the lower training time complexity than RankBoost but the identical test time complexity.

In this paper, the resource allocation problem with proportional fairness rate in cognitive OFDM-based wireless network is studied. It aims to maximize the total system throughput subject to constraints that include total transmit power for secondary users, maximum tolerable interferences of primary users, bit error rate, and proportional fairness rate among secondary users. It is a nonlinear optimization problem, for which obtaining the optimal solution is known to be NP-hard. An efficient bio-inspired suboptimal algorithm called immune clonal optimization is proposed to solve the resource allocation problem in two steps. That is, subcarriers are firstly allocated to secondary users assuming equal power assignment and then the power allocation is performed with an improved immune clonal algorithm. Suitable immune operators such as matrix encoding and adaptive mutation are designed for resource allocation problem. Simulation results show that the proposed algorithm achieves near-optimal throughput and more satisfying proportional fairness rate among secondary users with lower computational complexity.

In this paper, a method based on the Distribution of Character Skeleton is adopted to extract the structural features of handwriting image. In this method, we firstly extract the character skeleton by applying morphology and then compute the skeleton direction distribution in each sub-region as writing style logos of different writers. Comparing with Gabor texture analysis method, it demonstrates the feasibility and effectiveness of this method. We adopt Nearest neighbor classifier based on weighted, also the classification results verified the classification performance is better than Gabor texture analysis method and the correct identification rate is higher.

Grain is a national strategic resource, and is directly related to national security and social stability. The development of the grain internet of Things is highly valued by the government. The paper discusses the basic concept of internet of things, not only mentions the information function model of IOT for grain monitoring, but also proposes the architecture reference model of the internet of things for grain monitoring which is composed of four layers: perception layer, transport layer, processing layer and application layer. It also analyses the function of each layer, and discusses the key technical problems in future research.

Volatile oil contained in Atractylodes macrocephala, Zingiberis rhizoma, Costustroot and Radix glycyrrhizae, which were components in Wulian antidiarrheic granules, was extracted by distillation, the optimum extraction condition was determined by observing the imbitition amount, soaking time, extracting time and crashing degree, based on the dissolution ratio of volatile oil. Volatile oil was wrapped up with β-cyclodextrin (β-CD) to stabilize the preparation and ensure the therapeutic effect, the optimum clathration condition was investigated with orthogonal design by determining the yield rate of clathratio and availability of volatile oil. The results showed that the best extraction process of volatile oils was as following: Atractylodes macrocephala, Zingiberis rhizoma, Costustroot and Radix et rhizome glycyrrhizae were crushed into coarse powder, no soak, extracted 6 hours with 8 times amount of water; the clathration technology of volatile oil was: proportion of β-CD and volatile oil was 4:1, β-CD and water was 1:8, clathratio for 30 min at 40°C.

To address the parameter optimization problem of plate color recognition, two approaches based on IA (immune algorithm) and GA (genetic algorithm) are proposed respectively. Theoretical comparison of IA and GA is first made. Then experimental comparison of the two algorithms is given by using them to perform the parameter optimization task for color recognition of license plates. For plate color recognition algorithm, color features are extracted in the HSV (hue, saturation, and value) color space and weighted fusion of the fuzzy maps on three components is utilized to perform color recognition. To improve the adaptability of recognition algorithm, weights of color feature components and thresholds of classification functions are optimized by immune and genetic algorithms respectively. Comparison experiments were conducted on three data sets. And the experimental results show that the immune-based approach achieves higher accuracy and smaller mean square deviation. From the theoretical and experimental comparisons, it is shown that many immune mechanisms, such as clonal explosion, immune supplementation, concentration adjustment, etc. can be used to solve the parameter optimization problem effectively and efficiently.

To improve accuracy and adaptability, this paper presents a learning algorithm for color recognition of license plates. For three components of the hue-saturation-value (HSV) color space, different membership functions were defined to calculate their fuzzy degrees. Through the weighted fusion of the three membership degrees, a single map was produced to be the classification function for color recognition, and the final decision is based on the integrated map. Thresholds of membership functions, weight vectors of membership degrees and classification thresholds were all learned by the proposed learning algorithm, according to the classification error minimization inductive principle. Experiments were conducted on two different test sets. The overall accuracies of the proposed algorithm are 97.70% and 96.20%, respectively. The experimental results show that the proposed algorithm can learn the appropriate thresholds and weights from the training images, which are consistent with the practical application environments. Thus it improves the accuracy and adaptability of the color recognition algorithm and can meet the requirements of the practical engineering applications.

Considering the problem of feedback cost and performance loss for precoded spatial multiplexing MIMO OFDM with limited feedback, an improved nonlinear detection method based on perturbation codebook with limited feedback for precoded vblast is proposed, the corresponding criteria is used for select optimal codeword from the Grassmannian codebook at the receiving according to the channel state information, selecting optimal perturbation codeword, and feedback the Grassmannian precoding word index and perturbation codeword to the transmitter, thereby forming the equivalent channel. At the receiving, recycle and successive mapping from the entire space of constellation points as the first the underlying detection signals at the receiver, in order to maximize the performance of first detection layer, and then using the traditional detection algorithm for the remaining received signal, and determine a set of the optimal detection signal as decoding result based on the maximum likelihood criterion. Simulation results show that the proposed method to ensure system throughput, the cost of the feedback link and BER performance are improved some extent.

Building heterogeneous multi-core processor is desirable because it may provide high parallel performance such as instruction level parallelism (ILP) and thread-level parallelism (TLP). However, most existing operating system and parallel applications can't make good use of this heterogeneity because it arises dramatic instability sometimes, especially when the work of different core is unbalanced. In this paper we design a non-monotonic architecture consisted of two different types of cores to avoid the instability. Then we propose two strategies to ensure the load balance among different types of cores. In the end, we simulate this system and use these strategies to demonstrate our design is efficient. It alleviates the instability dramatically and the weighted speedup of heterogeneity-aware static scheduling (HASS) policy we examine outperforms straightforward scheduling (SS) policy by up to 2.4 and heterogeneity-aware dynamic thread migration (HADM) policy exceeds the SS policy by up to 3.

In this paper, to solve the problem of feedback cost and performance loss for precoded spatial multiplexing with limited feedback, an novel nonlinear detection method based on limited feedback for precoded vblast is proposed, the corresponding criteria is used for select optimal codeword from the Grassmannian codebook at the receiving according to the channel state information, and feedback the precoding word index to the transmitter, thereby forming the equivalent channel. At the receiving, recycle and successive mapping from the entire space of constellation points as the first the underlying detection signals at the receiver, in order to maximize the performance of first detection layer, and then using the traditional detection algorithm for the remaining received signal, and determine a set of the optimal detection signal as decoding result based on the maximum likelihood criterion. Simulation results show that the proposed method to ensure system throughput, the cost of the feedback link and BER performance are improved some extent.

Attribute importance ranking still is a key problem required to be solved for the classification problem. The lack of efficient heuristic information is the fundamental reason that affects the attribute selection in data mining. In this paper, to determining the importance level of the attributes, a new measure based on partial derivative distribution of the classification hypersurface output corresponding to the input attributes is proposed. This paper indicates that if the classification hypersurface is acquired by SVM, it is more convenient to measure the attribute importance ranking. The validity of the proposed method is experimentally evaluated, the experimental results prove that the approach is more efficient.

During the storage of grain, its quality can be influenced by temperature, humidity, oxygen, microorganisms, insects and other factors. It is necessary to protect the storage quality of grain through real-time monitoring these parameters and analyzing their influences. An algorithm is proposed based on flexible logic centralized fusion to study the evaluation method of overall temperature in grain heap. According to the temperature model of grain heap and its trends, grain quality can be evaluated and predicted.

Considering the problem of feedback cost and performance loss for precoded spatial multiplexing with limited feedback, an novel nonlinear detection method based on perturbation codebook with limited feedback for precoded vblast is proposed, the corresponding criteria is used for select optimal codeword from the Grassmannian codebook at the receiving according to the channel state information, selecting optimal perturbation codeword, and feedback the Grassmannian preceding word index and perturbation codeword to the transmitter, thereby forming the equivalent channel. At the receiving, recycle and successive mapping from the entire space of constellation points as the first the underlying detection signals at the receiver, in order to maximize the performance of first detection layer, and then using the traditional detection algorithm for the remaining received signal, and determine a set of the optimal detection signal as decoding result based on the maximum likelihood criterion. Simulation results show that the proposed method to ensure system throughput, the cost of the feedback link and BER performance are improved some extent.

Kernel principal component analysis (KPCA) is a popular nonlinear feature extraction method. It generally uses eigen-decomposition technique to extract the principal components in feature space. But the method is infeasible for large-scale data set because of the storage and computational problem. To overcome these disadvantages, an efficient iterative method of computing kernel principal components is proposed. First, the Power iteration is introduced to compute the first eigenvalue and corresponding eigenvector. Then the deflation method is repeatedly applied to achieve other higher order eigenvectors. In the process of computation, the kernel matrix needs not to compute and store in advance. The space and time complexity of the proposed method is greatly reduced. The effectiveness of proposed method is validated from experimental results.

A novel vision method is proposed to correct the robot heading angles in this paper. The salient feature point with known relative position to the target was chosen as the reference point. And the pose adjustment quantities of the robot were determined according to the navigation curve which is composed of the images of the reference point. This algorithm uses only one reference point and has the advantages of simple structure and low computation complexity. Experimental results show that the vision navigation method can eliminate the pose errors effectively, and the applicability and stability of the method are verified.

Color recognition of license plates is an important step to License Plate Recognition (LPR) system. In order to perform color recognition more effectively, an algorithm based on Naive Bayesian approach is proposed in this paper. To improve the efficiency of color recognition, the multiclass problem is converted into two binary problems based on the reverse color information of plate images. Color features of license plates are extracted in the HSV (hue-saturation-value) color space. Statistical information about color features for each type of license plates are obtained from training samples. And color classifier is built according to the Naive Bayesian rule. Comparison experiments were conducted on three different test sets. And the experimental results show that the proposed algorithm achieves good recognition performances.

A dual-polarization communication model with consideration of signal delay along receiving antennas in wireless sensor networks is proposed. Exact expressions for channel capacity and bit error rate are derived. It is manifested that the proposed dual-polarization model can improve channel capacity or reduce bit error rate under same data rate and total transmit power as compared with single antenna communication system if receive node is deployed appropriately. The dual-polarization model can significantly lower transmission power per bit requirement and prolong networks' lifetime. The robustness is displayed on small polarization angle.

We study a spectrum sharing scenario where multiple secondary systems coexist and interfere with each other. Unfair and inefficient situations may arise since negotiation is infeasible for competitive secondary users from dissimilar secondary systems. We use repeated games in microeconomic to construct cooperation due to lack of information exchange. Repeated games build reputations and apply punishments to achieve a desirable self-enforcing outcome. By playing a strategic game many times more efficient and fair Nash Equilibrium is acquired.

In creating a pattern classifier, feature selection is often used to prune irrelevant and noisy features to producing effective features. Manually developing a feature set can be a very time consuming and costly endeavor. In this paper, an efficient feature selection algorithm based on improved binary particle swarm optimization and support vector machine Algorithm (IBPSO-SVM) was used. First a population of particles (feature subsets) were randomly generated, and then optimized by IBPSO-SVM wrapper algorithms; finally the best fitness feature subset was applied to SVM classification. The simulation experiment results have proved that the feature subset selection algorithm based on IBPSO-SVM is very effective.

Grid system is inclined to support resource sharing for mixed types of workloads nowadays to improve the whole user utility, which requires an effective allocation method that not only each workload user can have desirable resources to satisfy its performance requirement from the view of user, but also can avoid shortage or waste of resources from the view of system. Currently, most resource allocation methods adopt economic/market models based on price policy that have two sides of deficiency, one is, price can only reflect supply and demand of one type of resource, and other is, competition among users is calculated but cooperation among users is ignored. We propose a resource allocation method constrained by a configured consumer sharing policy tree which can reflect not only competition but also cooperation relationship among users, and the ownership relationship between users and resources is also involved. The proposed allocation method that can support dynamic resource achievement of mixed types of workloads, such as batch workload, service workload etc., so that the application field of grid system is extended and the user utility of whole system is improved.

Resource management technology enables grid system to support on-demand resource achievement across multi organizations, which acquires great efforts in grid computing research field. Most current grid system can only serve one kind of workload for that resource management and workload management are tightly coupled, called system-centric resource management model, in which the resources utility for multi workloads is not much considered to have the problem of scarcity or waste of resources. We bring forward a multi-workloads sharing oriented resource management model for grids to have multi workloads built on one platform, in which multi workloads can utilize and share resources simultaneously and coordinately, so that the resource utility of whole grid system can be improved.

This paper proposes a method of document clustering algorithm based on Ant Colony Algorithm (ACO) and Fuzzy C-means Clustering (FCM). First, the algorithm makes use of the great ability of Ant Colony Algorithm for finding local extremum. It's derived from a basic model interpreting ant colony organization of cemeteries. The ACO Algorithm for flexibility, self-organization and robustness has been applied in a variety of areas. Taking advantage of these traits, good initial clusters are obtained at first step in our algorithm. Then, we combine these with Fuzzy C-means clustering organically. We also find out the whole distributing optimization clustering process, and achieve clustering analysis based on improved function. Experimental results show the good performance of the hybrid document clustering algorithm.

Boron doped hydrogenated nanocrystalline silicon (nc-Si:H) thin films are deposited using layer-by-layer technique in radio frequency (RF) plasma enhance chemical vapor deposition system. Studies about the influence of gas pressure, RF power density and buffer layer in i/p interface on the microstructure and electrical properties of Boron doped nc-Si:H thin films have been carried out. The experimental results have shown that a), as the gas pressure decreases, the grain size of nc-Si:H thin film increases, while its conductivity turns small, and the thin films transform into amorphous silicon when the gas pressure decreases to 100Pa; b), the microstructure of nc-Si:H thin films evolves to more ordered when the power density was enhanced from 280mW/cm2 to 560mW/cm2, and the grain size changes to be larger; c), the thickness of i/p buffer has large effect on the microstructure of Si:H thin films, when the deposition time of buffer layer increases from 0min to 10min, the Si:H thin films changes from amorphous to nanocrystalline, however the optimal time is about 2min. When the boron doped nc-Si:H thin films are applied as the window layer of n-i-p flexible amorphous silicon solar cells, comparing with p type a-SiC:H window layer, the performance of solar cells have been largely improved.

The lack of efficient heuristic information is the fundamental reason that causes the low effectiveness of currently used approaches for extracting symbolic rules. Classification hypersurface are direct heuristic information for selecting attributes. In SVM (support vector machine), classification hypersurface is emphasized because of the direct induction of the support vectors. In this paper, a new SVM model and a new measurement method of the classification power of attributes on the basis of the characteristics of the classification hypersurfaces based on the proposed SVM model is presented. The experimental results prove that the approach proposed can improve the validity of the extracted rules remarkably compared with other rule extracting approach.

The engineering experiments need to highlight the logics, steps, rules and details of involved intricate operations, but which are always neglected by the students. In this paper, we propose our workflow technology based virtual laboratory system (WFVL) for data mining. It subdivides the general data analysis model into detachable elements and encapsulates them as reusable Web services, which can be selected and linked referring to different operation logic or existing workflow model to create, achieve and observe different virtual experiments based on simulation data. WFVL uses the Predictive Model Markup Language (PMML) to describe different user models, which can be stored into model repository for teachers to create analysis or evaluation report of experimental teaching.

The amount of online document has grown greatly in recent years due to the increase in popularity of the World Wide Web. Thus, developing an efficient document classification method to automatically manipulate Web document is of great importance. A novel memetic algorithm (MA)-based document classification algorithm is presented in this paper. The experimental results show that the proposed algorithm achieves much better performance than other related classification algorithms.

Recent researches show that lung cancer owns actual dose-response relationship with calendar-year smoking environment exposure matrix and individual medical record. In this paper, two hybrid prediction models based on BP neural network, ES (exponential smoothing) and FCM (Fuzzy C-Means) clustering are proposed to predict the possible rate and ages of smokers suffering the lung cancer. The BP-ES (Exponential Smoothing) model can exert the superiorities of the time series datum of smoking crowds and other pathogenic factors; and the BPFCM clustering model can reduce the parameter amount and complexity of BP netpsilas training greatly. The experiments show that the accuracy of the hybrid models are enhanced greatly contrasted with single BP neural network, and can work as effective methods for the statistic, analysis and prediction to lung cancer.

To efficiently retrieve relevant documents from the explosive growth of the Internet and other sources of information access, an efficient document query optimization approach based on memetic algorithm (MA) is proposed. Experimental results show that the proposed algorithm can improve the precision of document retrieval compared with other conventional query optimization algorithm.

This paper proposes a digital watermarking for vector maps, some significant character strings can be embedded into vector maps. This algorithm based on data configuration of vector map, combines with the theory of error correction in the field of digital communication. Theoretic analysis and experimental results show that, the embedded watermarking induces less distortion, and it can resist some common manipulations, such as translation, rotation, scaling, vertex insertion/removal and local modification attacks, namely, it has perfectible robustness.

Qos multicast routing is of great importance for many multimedia communication applications such as videos conference, remote education, CSCW and so on. In order to effectively resolve Qos multicast routing, this paper mainly focus on the bandwidth-delay-constrained least-cost multicast routing problem, and present a memetic algorithm(MA) to deal with this problem. The experimental results show that the proposed algorithm achieves much better performance.

How to extract rules from trained SVMs has become an important preprocessing technique for data mining, pattern classification, and so on. There are two key problems required to be solved in the SVM based classification rule extraction, i.e. the attribute selection and the discretization to continuous attributes. In this paper, the differential characteristic of SVR (Support vector regression) is discussed. A new measure for determining the importance level of the attributes based on the trained SVR classifiers is proposed. Based on this new measure, a new approach for rule extraction from trained SVR classifiers is proposed. A new algorithm for rule extraction is given. The performance of the new approach is demonstrated by several computing cases. Experiment results show that the proposed approach can improve the validity of the extracted rules remarkably compared to other rule extracting approaches, especially for complicated classification problems.

SVM based rule extraction has become an important preprocessing technique for data mining, pattern classification, and so on. There are two key problems required to be solved in the classification rule extraction based on SVMs, i.e. the attribute importance ranking and the discretization to continuous attributes. In the paper, firstly, a new measure for determining the importance level of the attributes based on the trained SVR (Support vector re-gression) classifiers is proposed. Based on this new measure, a new approach for the division to continuous attribute space based on support vectors is pre-sented. A new approach for classification rule extraction from trained SVR classifiers is given. The performance of the new approach is demonstrated by several computing cases. The experimental results prove that the proposed ap-proach proposed can improve the validity of the extracted classification rules remarkably compared with other constructing rule approaches, especially for complicated classification problems.

Watermarking technology is the most efficient way to protect the ownership of multimedia data. In this paper, a novel image watermarking scheme using the Discrete Wavelet Transform (DWT) and memetic algorithm (MA) is introduced. The watermark is embedded to subband coefficients of subimage which is extracted from the original image by using DWT, and watermark extraction is efficiently performed via memetic algorithm. Experimental results show that the proposed watermarking scheme makes an almost invisible difference between the watermarked image and the original image, and is robust to common image processing operations.

To efficiently deal with Web document classification problem, a novel document classification algorithm based on shuffled frog leaping (SFL) algorithm is proposed in this paper. The SFL algorithm combines the benefits of the genetic-based memetic algorithms and the social behavior-based particle swarm optimization algorithms. The experimental results indicate that the proposed SFL algorithm yields much better performance than other conventional algorithms.

In this paper, we propose an Improved Ant Colony Optimization (IACO) and Hybrid Particle Swarm Optimization (HPSO) method for Spatial Clustering with Obstacles Constraints (SCOC). In the process of doing so, we first use IACO to obtain the shortest obstructed distance, and then we develop a novel HPKSCOC based on HPSO and K-Medoids to cluster spatial data with obstacles. The experimental results demonstrate that the proposed method, performs better than Improved K-Medoids SCOC in terms of quantization error and has higher constringency speed than Genetic K-Medoids SCOC.

The paper proposes a concept of embodying textual messages based on virtual or physical robotic avatar(s), to efficiently express non-verbal information embedded in contexts. With the concept, textual messages are expressed into speeches, facial expressions, and body languages through a robotic avatar. Interfaces constructed under such a concept are expected to have higher efficiency in exchanging information embedded in contexts than text-based interfaces do. A prototype based on virtual avatar(s) is developed. The avatar is constructed by regular surfaces. Body languages and some simple facial expressions are obtained by controlling movement of links and configuring body language functions. The system can generate one or multiple avatars for several intended applications.

Hydrogenated amorphous silicon thin film solar cells are prepared using radio frequency plasma enhanced chemical vapor deposition on polyimide substrates. The structure of the cell is polyimide /Al/n-μc-Si:H/i-a-Si:H/p-a-SiC:H/TCO. Due to amorphous silicon cells have Staebler-Wronski effect and their efficiencies do not stabilize at prepared, so these cells were placed more than fifteen months and I-V characteristics of these cells were measured in four times. They include (i)in the initial, (ii) placed in desiccator over fifteen months, (iii) light-soaked for five hours under AM 1.5 (100mW/cm 2 ) and (iv) annealed for two hours in natural conditions. The measurement results such as open-circuit voltage, short-circuit current, resistance in series, parallel resistance, conversion efficiency and fill factor are given and studied. Staebler-Wronski Effect are found and explained with these parameters. Generally, after light-soaking, the short-circuit currents of cells increase, the fill factors and open-circuit voltage decrease and the reduction of the efficiency is proportional to the fill factor. Abnormal Staebler-Wronski Effect appears in a cell. The cause of these phenomena is analyzed. The consequence that the efficiency varieties with the fill factor is derived r. A graded gap in P-I interface can be improved the cell efficiently. Ultimately, the stability of a-SiH thin film solar cells on polyimide substrate is discussed.

With advances in the computer technologies and the rapid development of Internet, information on the Internet is increasing exponentially. To efficiently retrieve relevant documents from the explosive growth of the Internet and other sources of information access, a novel Web document retrieval algorithm based on particle swarm optimization (PSO) and linear discriminant analysis (LDA) algorithm is proposed to deal this problem. Experimental results clearly demonstrate its effectiveness and efficiency.

In this paper, a novel blind watermark extracting scheme using the Discrete Wavelet Transform (DWT) and Particle Swarm Optimization (PSO) algorithm is introduced. The watermark is embedded to the discrete multiwavelet transform (DMT) coefficients larger than some threshold values, and watermark extraction is efficiently performed via particle swarm optimization algorithm.The experimental results show that the proposed watermarking scheme results in an almost invisible difference between the watermarked image and the original image, and is robust to common image processing operations and JPEG lossy compression.

Intelligent visual surveillance system can work normally under clear weather. But under bad weather, especially in foggy days, it can not detect moving objects accurately due to low scene visibility. Our research aims to resolve this problem. This paper presents a novel method for moving object detection in foggy days. Firstly, surveillance video under foggy weather is defogged, leveraging a physics-based image restoration approach. Secondly, we exploit a novel background maintenance algorithm based on the Unscented Kalman Filter(UKF) to subtract the background from the defogged video. Finally, moving objects are segmented by background differencing. Evaluations are performed to verify the effectiveness and practicality of this approach. Experimental results show that our method can be applied in real time surveillance systems.

Due to the exponential growth of documents in the Internet and the emergent need to organize them, the automatic document classification has received an ever-increased attention in the recent years. The particle swarm optimization (PSO) algorithm, new to the document classification community, is a robust stochastic evolutionary algorithm based on the movement and intelligence of swarms. In this paper, a PSO-based algorithm for document classification is presented. Comparison between our method and other conventional document classification algorithms is conducted on Reuter and TREC corpora. The experimental results indicate that our proposed algorithm yields much better performance than other conventional algorithms.

Support vector regression (SVR) is a new technique for pattern classification , function approximation and so on. In this paper we propose an new constructing approach of classification rules based on support vector regression and its derivative characteristics for the classification task of data mining. a new measure for determining the importance level of the attributes based on the trained SVR is proposed. Based on this new measure, a new approach for clas-sification rule construction using trained SVR is proposed. The performance of the new approach is demonstrated by several computing cases. The experimen-tal results prove that the approach proposed can improve the validity of the ex-tracted classification rules remarkably compared with other constructing rule approaches, especially for the complicated classification problems.

The particle swarm optimization (PSO) algorithm, is a robust stochastic evolutionary algorithm based on the movement and intelligence of swarms. In this paper, a PSO-based algorithm for color image quantization is presented. Compared with other well-known color image quantization algorithm in commonly used images, the proposed algorithm is effective and achieves a quite fast executing speed.

Data dimensionality reduction(DDR) is an important preprocessing technique for data mining, pattern classification and so on. DDR aims at obtaining compact representation of the original data while reduce unimportant or irrelevant data. In this paper we propose a new measure for determining the importance level of the attributes based on the trained support vector regression (SVR) and its derivative characteristics. Based on this new measure, a new approach for data dimensionality reduction based on support vector regression is proposed. The performance of the new approach is demonstrated by several computing cases. The experimental results prove that the approach proposed can improve efficiency and effectiveness significantly compared with other data dimensionality reduction approaches.

Classification rule mining is one of the important problems in the emerging field of data mining which is aimed at finding a small set of rules from the training data set with predetermined targets. To efficiently mine the classification rule from databases, a novel classification rule mining algorithm based on particle swarm optimization (PSO) was proposed. The experimental results show that the proposed algorithm achieved higher predictive accuracy and much smaller rule list than other classification algorithm.

Text Categorization is the process of assigning documents to a set of previously fixed categories. In this paper we develop an optimal SVM algorithm for text classification via multiple optimal strategies, such as a novel importance weight definition, the feature selection using the likelihood ratio for binomial distribution, the optimal parameter settings, etc. Comparison between our method and other conventional text classification algorithms is conducted on Reuter and TREC corpora. The experimental results indicate that our proposed algorithm yields much better performance than other conventional algorithms.

Text classification is a problem of assigning a document into one or more predefined classes. One of the most interesting issues in text categorization is feature selection. This paper proposes a novel approach in feature selection based on support vector machine(SVM) and latent semantic indexing(LSI), which can identify LSI-subspace that is suited for classification. Experimental results show that the proposed method can achieve higher classification accuracies and is of less training and prediction time.

The strong nonlinear relation between the training sample's impact on the errors and error's derivatives is the fundamental reason underlying the low learning efficiency of the multi-layer forward neural networks. Effectively decreasing the degree of the nonlinear relation and its impact on network learning is critical to improve the neural network's training efficiency. Based on the above idea, this paper propose a new approach to accelerate learning efficiency, including linearization technique of the non-linear relation, the convergence technique based on the local equalization of training sample's errors, and the rotation adjustment of the weights. A new fast learning algorithm for the multi-layer forward neural networks is also presented. The experimental results prove that the new algorithm is capable of shortening the training time by hundreds time and remarkably improve generalization of the neural networks, compared with the conventional algorithms.