Abstract: Suppose a decision maker has to purchase a commodity over time with varying prices and demands. In particular, the price per unit might depend on the amount purchased and this price function might vary from step to step. The decision maker has a buffer of bounded size for storing units of the commodity that can be used to satisfy demands at later points in time. We seek for an algorithm deciding at which time to buy which amount of the commodity so as to minimize the cost. This kind of problem arises in many technological and economical settings like, e.g., battery management in hybrid cars and economical caching policies for mobile devices. A simplified but illustrative example is a frugal car driver thinking about at which occasion to buy which amount of gasoline. Within a regret analysis, we assume that the decision maker can observe the performance of a set of expert strategies over time and synthesizes the observed strategies into a new online algorithm. In particular, we investigate the external regret obtained by the well-known Randomized Weighted Majority algorithm applied to our problem. We show that this algorithm does not achieve a reasonable regret bound if its random choices are independent from step to step, that is, the regret for T steps is Ω(T ). However, one can achieve regret O( √ T ) when introducing dependencies in order to reduce the number of changes between the chosen experts. If the price functions satisfy a convexity condition then one can even derive a deterministic variant of this algorithm achieving regret O( √ T ). Our more detailed bounds on the regret depend on the buffer size and the number of available experts. The upper bounds are complemented by a matching lower bound on the best possible external regret. ∗Supported by the DFG GK/1298 “AlgoSyn” and UMIC Research Center

Abstract: Suppose a decision maker has to purchase a commodity over time with varying prices and demands. In particular, the price per unit might depend on the amount purchased and this price function might vary from step to step. The decision maker has a buffer of bounded size for storing units of the commodity that can be used to satisfy demands at later points in time. We seek for an algorithm deciding at which time to buy which amount of the commodity so as to minimize the cost. This kind of problem arises in many technological and economical settings like, e.g., battery management in hybrid cars and economical caching policies for mobile devices. A simplified but illustrative example is a frugal car driver thinking about at which occasion to buy which amount of gasoline. Within a regret analysis, we assume that the decision maker can observe the performance of a set of expert strategies over time and synthesizes the observed strategies into a new online algorithm. In particular, we investigate the external regret obtained by the well-known Randomized Weighted Majority algorithm applied to our problem. We show that this algorithm does not achieve a reasonable regret bound if its random choices are independent from step to step, that is, the regret for T steps is Ω(T ). However, one can achieve regret O( √ T ) when introducing dependencies in order to reduce the number of changes between the chosen experts. If the price functions satisfy a convexity condition then one can even derive a deterministic variant of this algorithm achieving regret O( √ T ). Our more detailed bounds on the regret depend on the buffer size and the number of available experts. The upper bounds are complemented by a matching lower bound on the best possible external regret. ∗Supported by the DFG GK/1298 “AlgoSyn” and UMIC Research Center

Abstract: Decision tree algorithm is a kind of data mining model to make induction learning algorithm based on examples. It is easy to extract display rule, has smaller computation amount, and could display important decision property and own higher classification precision. For the study of data mining algorithm based on decision tree, this article put forward specific solution for the problems of property value vacancy, multiple-valued property selection, property selection criteria, propose to introduce weighted and simplified entropy into decision tree algorithm so as to achieve the improvement of ID3 algorithm. The experimental results show that the improved algorithm is better than widely used ID3 algorithm at present on overall performance.

Abstract: This paper presents the G3PARM algorithm for mining representative association rules. G3PARM is an evolutionary algorithm that uses G3P (Grammar Guided Genetic Programming) and an auxiliary population made up of its best individuals who will then act as parents for the next generation. Due to the nature of G3P, the G3PARM algorithm allows us to obtain valid individuals by defining them through a context-free grammar and, furthermore, this algorithm is generic with respect to data type. We compare our algorithm to two multiobjective algorithms frequently used in literature and known as NSGA2 (Non dominated Sort Genetic Algorithm) and SPEA2 (Strength Pareto Evolutionary Algorithm) and demonstrate the efficiency of our algorithm in terms of running-time, coverage and average support, providing the user with high representative rules.

Abstract: The AdaBoost.M1 is one of the machine learning algorithms. But it will fail if the weak learner cannot achieve at least 50% accuracy when run on these hard distributions. Random Forest is computationally effective and offer good prediction performance. A new approach AdaBoost.M1-RF algorithm, which using Random Forest as weak learner, is proposed in the paper. To evaluate the performance of AdaBoost.M1-RF algorithm, it is compared with other machine learning algorithms.

Abstract: Summary The effectiveness of wireless sensor networks (WSNs) depends on the coverage and target detection probability provided by dynamic deployment, which is supported by the several methods. Particle Swarm Optimization (PSO) algorithm is one these methods, however computation time required is a big bottleneck. This paper proposes three dynamic PSO-based deployment algorithms that reduce the computation time. First algorithm by the name of “PSO-LA” algorithm comprised of PSO algorithm and learning automata. In this algorithm, speed of particles is corrected by using the existing knowledge and the feedback from the actual implementation of the algorithm. Hence in this algorithm mobile nodes move more objectively than PSO and achieve the result with less number of repetitions. To improve performance of this algorithm, second algorithm by the name of “Improved PSO-LA” algorithm is introduced, regulating its movement without an impact from the movement of other mobile nodes and based on the result gained from its previous movement. The first and second algorithms require sensors to move iteratively, eventually reaching the final destination. In the third algorithm by the name of “Improved PSO-LA with logical movement” with the same round-by-round procedure of the second algorithm, sensors calculate their target locations, virtually move there. The real movement only happens at the last round after final destinations are determined. Simulation results show the effectiveness of our proposed algorithms against other common approaches like VF and PSO algorithms.

Abstract: The automatic tuning of the parameters of algorithms and automatic selection of algorithms has received a lot of attention recently. One possible approach is the use of machine learning techniques to learn classifiers which, given the characteristics of a particular problem, make a decision as to which algorithm or what parameters to use. Little research has been done into which machine learning algorithms are suitable and the impact of picking the "right" over the "wrong" technique. This paper investigates the differences in performance of several techniques on different data sets. It furthermore provides evidence that by using a meta-technique which combines several machine learning algorithms, we can avoid the problem of having to pick the "best" one and still achieve good performance.

Abstract: In this paper, a new multi-objective reinforcement learning algorithm for multi-objective sequential decision making problems in unknown environment is proposed. The salient characters of the algorithm are: 1) decision maker's objective preference is introduced to guide learning direction; 2) a new measure of comparing action decisions under several objectives based on the fuzzy inference system is defined; 3) fast learning speed can be achieved. Simulation results demonstrate that the proposed algorithm has a good learning performance.

Abstract: An approximate algorithm for minimization of weighted depth of decision trees is considered. A bound on accuracy of this algorithm is obtained which is unimprovable in general case. Under some natural assumptions on the class NP, the considered algorithm is close (from the point of view of accuracy) to best polynomial approximate algorithms for minimization of weighted depth of decision trees.

Abstract: The famous F5 algorithm for computing Grobner basis was presented by Faugere in 2002 without complete proofs for its correctness. The current authors have simplified the original F5 algorithm into an F5 algorithm in Buchberger's style (F5B algorithm), which is equivalent to original F5 algorithm and may deduce some F5-like versions. In this paper, the F5B algorithm is briefly revisited and a new complete proof for the correctness of F5B algorithm is proposed. This new proof is not limited to homogeneous systems and does not depend on the strategy of selecting critical pairs (i.e. the strategy deciding which critical pair is computed first) such that any strategy could be utilized in F5B (F5) algorithm. From this new proof, we find that the special reduction procedure (F5-reduction) is the key of F5 algorithm, so maintaining this special reduction, various variation algorithms become available. A natural variation of F5 algorithm, which transforms original F5 algorithm to a non-incremental algorithm, is presented and proved in this paper as well. This natural variation has been implemented over the Boolean ring. The two revised criteria in this natural variation are also able to reject almost all unnecessary computations and few polynomials reduce to 0 in most examples.

Abstract: A new associative classification algorithm based on weighted voting (ACWV) is presented. ACWV takes advantage of two methods: the optimal rule method preferring high-quality rules and the voting method considering the majority of the rules. Moreover, the method takes into account both the length and convictions of rules to calculate their weights. First, ACWV builds a class-count FP-tree (called CCFP-tree) from the given historical data. After that, the weighted voting result for a new instance can be obtained from the CCFP-tree directly without storing, retrieving and sorting rules explicitly. The label of the class with maximal sum of weighted votes is then that of the new instance. Results of the experiments with 36 data sets selected from the UCI machine learning repository show that the proposed method has its advantages in comparison with previous methods in terms of classification accuracy.

Abstract: In traditional machine learning applications, only labeled data is used to train the classifier. Labeled data are difficult, expensive, time-consuming and require human experts to be obtained in several real applications. Semi-supervised learning address this issue. Semi-supervised learning uses large amount of unlabeled data, combined with the labeled data, to build better classifiers. The semi-supervised algorithm could be an extension of an unsupervised algorithm. Such algorithm would be based on unsupervised clustering algorithms, adding a term in its objective function that makes use of labeled information to guide the learning process. This study presents a new algorithm for semi-supervised clustering based on Fuzzy C-Means algorithm. The classifier was evaluated and compared against two semi-supervised clustering algorithms in the context of learning from partially labeled data. The behavior of the proposed algorithm is discussed and the results are validated using cross-validation and the confidence interval. Thus, it was possible to certify the better accuracy performance of the new algorithm when a few labeled data are available.

Abstract: We consider a repeated pricing decision problem of a monopolist (the decision-maker) who does not know the demand function of some new product, and hence the profit function. To decide, she is helped by a committee of N experts. Each expert has an estimation of the unknown demand function and use it to advise the decision-maker on how she should modify the current price. Decisions are taken with a weighted majority rule, where the weight of each expert, which may be interpreted as her decision power, evolves as a function of its accuracy. When a perfect exists, i.e., who always gives the correct advice, we show that she ends up with all the decision power in the long-run so that the decision-maker finds the optimal price. When such a perfect does not exist, the decision-maker is actually unable to consistently select an expert over time so that the sequences of prices and weights describe a limit cycle. Interestingly enough, if the decision-maker takes a large sample of the stationary behavior of prices, the empirical mean turns out to be arbitrarily close to the optimal price, independently of the "quality" of the experts, as long as there experts are "diverse" enough. This result gives thus support to the thesis developed in the books of Surowiecki (The wisdom of the crowd. Why the many are smarter than the few and how collective wisdom shapes business, economies, societies and nations, 2005) and Page (The difference: How the power of diversity creates better groups, firms, schools, and societies, 2007).

Abstract: Pattern matching is one of the important issues in the areas of network security and many others. The increase in network speed and traffic may cause the existing algorithms to become a performance bottleneck. Therefore, it is very necessary to develop more efficient pattern matching algorithm, in order to overcome troubles on performance. There are several algorithms in use, in which, some are with different methodology and other are with the improved methodology for the existing algorithms. In this paper, we are proposing a novel pattern matching algorithm, called, DP algorithm (Devaki - Paul algorithm). The algorithm works basing on some novel set of innovated rules, which will endorse the algorithm resulting in better performance and efficiency. In case of unsuccessful search, the DP algorithm has zero character comparisons, irrespective of the sizes of the text and pattern, provided if either the first or the last character was not present in the given input text. Whereas, the Booyer-Moore and Quick Search algorithms will do search as usual. The algorithm also doesn't require any pre-processing phase, if the search is on the same given input text and with different patterns, provided the first and the last characters are same as in the case of first pattern. The algorithm was tested and validated and the results have proved that the performance of DP algorithm is better than BM algorithm (Boyer - Moore algorithm) and Quick Search algorithm. In case of tests with repeated character, its performance is greater than 1%~50% with BM and Quick Search algorithms. In case of tests with the English Text and Random Pattern, it's greater than 33%~91% with BM and 37%~85% with Quick Search algorithms. In case of tests with the English Text and Random Pattern of an unsuccessful search, its performance is greater than BM and Quick Search algorithms with 100%, if either the first and/or the last character of the pattern in the given text were not present.

Abstract: A new fuzzy weighted average computation algorithm (NFWA) based on the α-cuts representation of fuzzy numbers is presented in this paper. For each α-cuts, the endpoints of the fuzzy weighted average (FWA) can be calculated from two particular switch points. In the proposed algorithm, these two switch points are computed with an opposite direction searching process, although recursive, which is remarkably efficient. The calculation complexity of the new algorithm is O(n). Experimental result demonstrates that compared with some commonly used FWA algorithms, the new algorithm approach requires the least CPU time, and then may be the fastest available FWA algorithm to date.

Abstract: Naive Bayes algorithm is a simple and efficient classification algorithm, but its conditional independence assumption is not always true in real life which is affected to some extent. Weighted Naive Bayesian classifier relax the conditional independence assumption to increase accuracy. Based on Identifiably matrix of Rough Set, a new weighted naive Bayes method based on attribute frequency is proposed. Different condition attributes are weighted differently; the Naive Bayesian classification algorithm performance is improved effectively. Experiments have proved that the calculation of this algorithm is easier and more effective.

Abstract: We propose algorithms for producing weighted majority votes that learn by probing the empirical risk of a randomized (uniformly weighted) majority vote--instead of probing the zero-one loss, at some margin level, of the deterministic weighted majority vote as it is often proposed. The learning algorithms minimize a risk bound which is convex in the weights. Our numerical results indicate that learners producing a weighted majority vote based on the empirical risk of the randomized majority vote at some finite margin have no significant advantage over learners that achieve this same task based on the empirical risk at zero margin. We also find that it is sufficient for learners to minimize only the empirical risk of the randomized majority vote at a fixed number of voters without considering explicitly the entropy of the distribution of voters. Finally, our extensive numerical results indicate that the proposed learning algorithms are producing weighted majority votes that generally compare favorably to those produced by AdaBoost.

Abstract: In dealing with a large number of train samples, Support Vector Regression (SVR) algorithm is slow. In particular, while new sample is added, all the training samples must be re-trained. In this paper, a new SVR incremental algorithm is presented, which is based on boundary vector. The algorithm takes full advantages of the geometric information of training sample sets. The observed data of China's GDP is used as a case study for the new algorithm. The computing results show that the new algorithm not only can guarantee the accuracy of machine learning and good generalization ability, but also can increase the learning speed of the algorithm than the classical SVR algorithm, and can be used rapid incremental learning.

Abstract: We consider a repeated pricing decision problem of a monopolist (the decision-maker) who does not know the demand function of some new product, and hence the profit function. To decide, she is helped by a committee of N experts. Each expert has an estimation of the unknown demand function and use it to advise the decision-maker on how she should modify the current price. Decisions are taken with a weighted majority rule, where the weight of each expert, which may be interpreted as her decision power, evolves as a function of its accuracy. When a perfect exists, i.e., who always gives the correct advice, we show that she ends up with all the decision power in the long-run so that the decision-maker finds the optimal price. When such a perfect does not exist, the decision-maker is actually unable to consistently select an expert over time so that the sequences of prices and weights describe a limit cycle. Interestingly enough, if the decision-maker takes a large sample of the stationary behavior of prices, the empirical mean turns out to be arbitrarily close to the optimal price, independently of the "quality" of the experts, as long as there experts are "diverse" enough. This result gives thus support to the thesis developed in the books of Surowiecki (The wisdom of the crowd. Why the many are smarter than the few and how collective wisdom shapes business, economies, societies and nations, 2005) and Page (The difference: How the power of diversity creates better groups, firms, schools, and societies, 2007).

Abstract: Selective ensemble learning is a learning algorithm, trains a number of based classifier and selects some of them to ensemble. Through the selective ensemble, the algorithm would be more effective than each single one and better than the algorithm that select all the based classifier, and the algorithm would have effective generalization ability. In this paper, we apply a multi-lever selective ensemble learning algorithm, named. Test results show that has better learning results and generalization ability.

Abstract: Traditional machine learning and data mining algorithms mainly assume that the training and test data must be in the same feature space and follow the same distribution. However, in real applications, these two hypotheses are difficult to hold, traditional algorithms are hence no longer applicable. As a new framework of learning, transfer learning could solve this problem effectively. This paper focuses on one of important branches in this field, namely inductive transfer learning. Correspondingly, a weighted algorithm of inductive transfer learning, based on maximum entropy model, is proposed, called WTLME. It transfers the model parameters learned from the source domain to the target domain, and meanwhile adjusts the weights of instances in the target domain to obtain the model with high accuracy. Extensive studies demonstrate that our proposed algorithm of WTLME is more effective and efficient than traditional algorithms that require learning from scratch if the data distributions change. Moreover, WTLME is comparable to the previous transfer algorithm based on maximum entropy model.

Abstract: In this paper, we propose a sparse semi-supervised learning method, which combines the large margin approach and the L1 constraint. The main contribution of the paper is to develop an efficient optimization algorithm. The objective function to be minimized in a large margin semi-supervised learning method is non-convex and non-differentiable, and hence special optimization algorithms are required. For this purpose, we develop an optimization algorithm, which is a hybrid of the CCCP and the gradient LASSO algorithm. The advantage of the proposed method over existing semi-supervised learning methods is that it can identify a small number of relevant input variables while keeping the prediction accuracy high. Also, the proposed algorithm is simple enough that it can be applied to various real problems without being much hampered by computational limitations. To confirm these advantages, we compare the proposed method with the standard semi-supervised method by analyzing simulated as well as real data sets.

Abstract: This paper improves CNM algorithm to detect community structure on weighted network. Based on the link weight and vertex weight, algorithm design defines a new Q-function to calculate community modularity, the type of communities were classified by finding the Q peak. We have generated networks with known community structure A,B and C(different sizes), to test if the algorithms can recognize and extract this structure. The results show that our algorithms perform well .The accuracy of detecting community structures shows feasibility and replicability of the algorithm.

Abstract: We extend the Chow-Liu algorithm for general random variables while the previous versions only considered finite cases. In particular, this paper applies the generalization to Suzuki's learning algorithm that generates from data forests rather than trees based on the minimum description length by balancing the fitness of the data to the forest and the simplicity of the forest. As a result, we successfully obtain an algorithm when both of the Gaussian and finite random variables are present.

Abstract: The least squares support vector machine (LS-SVM) is an interesting variant of the SVM. It performs structural risk through margin-maximization and has excellent power of generalization. For some applications, it is more interesting to use the weighted LS-SVM where the impact of each training sample is controlled by weighting factors. In this paper, we consider the use of the weighted LS-SVM in semi-supervised learning. We propose an algorithm to perform this type of learning by extending the transductive SVM idea. We tested our algorithm on both artificial and real problems and demonstrate its usefulness comparing with other semi-supervised learning methods.

Abstract: Weighted association rule algorithm for less than two: one does not meet the requirement of closed down,that is a subset of frequent sets may not be frequent sets;another weighted association rules can not handle the different importance of different items,and truly embody the the importance of different items different.In this paper,the dynamic weighted association rules algorithm,algorithm based on the importance of the project and the largest number of frequent itemsets at different stages of the project to determine the different weights,fully reflect the importance of different items are different,and algorithms are proven closed down.Experimental results show that the algorithm improves the high accuracy and efficiency.

Abstract: We have studied the AC-4 algorithm and then present value ordering heuristic for solving algorithm BT-MSV which is based on the AC-4 algorithm. This algorithm takes full advantage of supported information which is recorded in the data structure of the AC-4 algorithm during the process of arc consistency. The algorithm sorts the value of the variables' domain according to the supported information. So this order enforces the algorithm to extend the values of variables which have more support. In this way, the efficiency of the algorithm can be improved. The result of our experiments shows that our algorithm has much more advantage over other solving algorithms.