Nonlinear system identification using optimized dynamic neural network

In this paper, both off-line architecture optimization and on-line adaptation have been developed for a dynamic neural network (DNN) in nonlinear system identification. In the off-line architecture optimization, a new effective encoding scheme—Direct Matrix Mapping Encoding (DMME) method is proposed to represent the structure of neural network by establishing connection matrices. A series of GA operations are applied to the connection matrices to find the optimal number of neurons on each hidden layer and interconnection between two neighboring layers of DNN. The hybrid training is adopted to evolve the architecture, and to tune the weights and input delays of DNN by combining GA with the modified adaptation laws. The modified adaptation laws are subsequently used to tune the input time delays, weights and linear parameters in the optimized DNN-based model in on-line nonlinear system identification. The effectiveness of the architecture optimization and adaptation is extensively tested by means of two nonlinear system identification examples.     

Automatic detection of breast cancers in mammograms using structured support vector machines

Breast cancer is one of the most common cancers diagnosed in women. Large margin classifiers like the support vector machine (SVM) have been reported effective in computer-assisted diagnosis systems for breast cancers. However, since the separating hyperplane determination exclusively relies on support vectors, the SVM is essentially a local classifier and its performance can be further improved. In this work, we introduce a structured SVM model to determine if each mammographic region is normal or cancerous by considering the cluster structures in the training set. The optimization problem in this new model can be solved efficiently by being formulated as one second order cone programming problem. Experimental evaluation is performed on the Digital Database for Screening Mammography (DDSM) dataset. Various types of features, including curvilinear features, texture features, Gabor features, and multi-resolution features, are extracted from the sample images. We then select the salient features using the recursive feature elimination algorithm. The structured SVM achieves better detection performance compared with a well-tested SVM classifier in terms of the area under the ROC curve.     

Low cost inductor current sensing for power converters using feedback compensation

One of the key components in the current mode control of power converters is the sensing of inductor current. Traditional low cost implementation is to add a sensing resistor in series with the inductor to sense the inductor current. This will reduce the overall efficiency of the power converter. An alternative is to apply a current transformer to the inductor to sense its current. This however will significantly increase the cost of power converters. One low cost implementation is to add a resistor–capacitor network in parallel with the inductor so that the inductor current can be derived from the capacitor voltage. The problem of this implementation is that the capacitor voltage is very sensitive to temperature variations. In this paper, a simple feedback loop is applied to the sensed capacitor voltage based on the steady state duty ratio of power converters such that the sensed inductor current will be independent of temperature variations. The sensing element with feedback compensation can be applied to inductors fitted to power converters such as buck converters and boost converters. Although experimental studies based on a buck converter were carried out to illustrate the correctness of the new sensing technique, the proposed technique is applicable to other converter topologies.     

Modeling Default Data Via an Interactive Hidden Markov Model

In this paper, we first introduce the use of an interactive hidden Markov model (IHMM) for modeling and analyzing default data in a sector. Under the IHMM, transitions of the hidden risk states of the sector depend on the observed number of bonds in the sector that default in the current time period. This incorporates the feedback effect of the number of defaults on the transitions of the hidden risk states. This feature seems to be more realistic and does not enjoy by the traditional HMMs. We then develop a “dynamic” version of the binomial expansion technique (BET) modulated by the IHMM for modeling the occurrence of defaults of bonds issued by firms in the same sector. Under the BET modulated by the IHMM, the number of bonds defaulting in each time period follows a Markov-modulated binomial distribution with the probability of defaulting of each bond depending on the states of the IHMM, which represent the hidden risk states of the sector. Efficient method will be presented for estimating the model parameters in the BET modulated by the IHMM. We shall compare the hidden risk state process extracted from the IHMM-modulated BET with that extracted from the BET modulated by HMM in order to illustrate the significance of the feedback effect using real data. We shall also present the estimation results for the BET modulated by the IHMM and compare them with those for the BET modulated by the HMM.     

Effects of the number of hidden nodes used in a structured-based neural network on the reliability of image classification

A structured-based neural network (NN) with backpropagation through structure (BPTS) algorithm is conducted for image classification in organizing a large image database, which is a challenging problem under investigation. Many factors can affect the results of image classification. One of the most important factors is the architecture of a NN, which consists of input layer, hidden layer and output layer. In this study, only the numbers of nodes in hidden layer (hidden nodes) of a NN are considered. Other factors are kept unchanged. Two groups of experiments including 2,940 images in each group are used for the analysis. The assessment of the effects for the first group is carried out with features described by image intensities, and, the second group uses features described by wavelet coefficients. Experimental results demonstrate that the effects of the numbers of hidden nodes on the reliability of classification are significant and non-linear. When the number of hidden nodes is 17, the classification rate on training set is up to 95%, and arrives at 90% on the testing set. The results indicate that 17 is an appropriate choice for the number of hidden nodes for the image classification when a structured-based NN with BPTS algorithm is applied.     

Agent-based negotiation and decision making for dynamic supply chain formation

Modern businesses are facing the challenge of effectively coordinating their supply chains from upstream to downstream services. It is a complex problem to search, schedule, and coordinate a set of services from a large number of service resources under various constraints and uncertainties. Existing approaches to this problem have relied on complete information regarding service requirements and resources, without adequately addressing the dynamics and uncertainties of the environments. The real-world situations are complicated as a result of ambiguity in the requirements of the services, the uncertainty of solutions from service providers, and the interdependencies among the services to be composed. This paper investigates the complexity of supply chain formation and proposes an agent-mediated coordination approach. Each agent works as a broker for each service type, dedicated to selecting solutions for each service as well as interacting with other agents in refining the decision making to achieve compatibility among the solutions. The coordination among agents concerns decision making at strategic, tactical, and operational level. At the strategic level, agents communicate and negotiate for supply chain formation; at the tactical level, argumentation is used by agents to communicate and understand the preferences and constraints of each other; at the operational level, different strategies are used for selecting the preferences. Based on this approach, a prototype has been implemented with simulated experiments highlighting the effectiveness of the approach.     

Immunity-based hybrid evolutionary algorithm for multi-objective optimization in global container repositioning

The development of evolutionary algorithms for optimization has always been a stimulating and growing research area with an increasing demand in using them to solve complex industrial optimization problems. A novel immunity-based hybrid evolutionary algorithm known as Hybrid Artificial Immune Systems (HAIS) for solving both unconstrained and constrained multi-objective optimization problems is developed in this research. The algorithm adopts the clonal selection and immune suppression theories, with a sorting scheme featuring uniform crossover, multi-point mutation, non-dominance and crowding distance sorting to attain the Pareto optimal front in an efficient manner. The proposed algorithm was verified with nine benchmarking functions on its global optimal search ability as well as compared with four optimization algorithms to assess its diversity and spread. Sensitivity analysis was also carried out to investigate the selection of key parameters of the algorithm. It is found that the developed immunity-based hybrid evolutionary algorithm provides a useful means for solving optimization problems and has successfully applied to the problem of global repositioning of containers, which is one of a constrained multi-objective optimization problem. The developed HAIS will assist shipping liners on timely decision making and planning of container repositioning operations in global container transportation business in an optimized and cost effective manner.     

Ontology-based data mining approach implemented for sport marketing

Since sport marketing is a commercial activity, precise customer and marketing segmentation must be investigated frequently and it would help to know the sport market after a specific customer profile, segmentation, or pattern come with marketing activities has found. Such knowledge would not only help sport firms, but would also contribute to the broader field of sport customer behavior and marketing. This paper proposes using the Apriori algorithm of association rules, and clustering analysis based on an ontology-based data mining approach, for mining customer knowledge from the database. Knowledge extracted from data mining results is illustrated as knowledge patterns, rules, and maps in order to propose suggestions and solutions to the case firm, Taiwan Adidas, for possible product promotion and sport marketing.     

A new document representation using term frequency and vectorized graph connectionists with application to document retrieval

This paper presents a new document representation with vectorized multiple features including term frequency and term-connection-frequency. A document is represented by undirected and directed graph, respectively. Then terms and vectorized graph connectionists are extracted from the graphs by employing several feature extraction methods. This hybrid document feature representation more accurately reflects the underlying semantics that are difficult to achieve from the currently used term histograms, and it facilitates the matching of complex graph. In application level, we develop a document retrieval system based on self-organizing map (SOM) to speed up the retrieval process. We perform extensive experimental verification, and the results suggest that the proposed method is computationally efficient and accurate for document retrieval.     

Imbalanced text classification: A term weighting approach

The natural distribution of textual data used in text classification is often imbalanced. Categories with fewer examples are under-represented and their classifiers often perform far below satisfactory. We tackle this problem using a simple probability based term weighting scheme to better distinguish documents in minor categories. This new scheme directly utilizes two critical information ratios, i.e. relevance indicators. Such relevance indicators are nicely supported by probability estimates which embody the category membership. Our experimental study using both Support Vector Machines and Naïve Bayes classifiers and extensive comparison with other classic weighting schemes over two benchmarking data sets, including Reuters-21578, shows significant improvement for minor categories, while the performance for major categories are not jeopardized. Our approach has suggested a simple and effective solution to boost the performance of text classification over skewed data sets.