Catálogo de publicaciones - libros

A competitive neural network provides a highly effective means of attaining a sound solution and of reducing the network complexity. A competitive approach is utilized to deal with fully-utilized scheduling problems. This investigation employs slack competitive Hopfield neural network (SCHNN) to resolve non-fully and fully utilized identical machine scheduling problems with multi-constraint, real time (execution time and deadline constraints) and resource constraints. To facilitate resolving the scheduling problems, extra slack neurons are added on to the neural networks to represent pseudo-jobs. This study presents an energy function corresponding to a neural network containing slack neurons. Simulation results demonstrate that the proposed energy function integrating competitive neural network with slack neurons can solve fully and non-fully utilized real-time scheduling problems.

The process scheduling aims to arrange CPU time to multiple processes for providing users with more efficient throughput. Except the class of process set by user, conventional operating systems have applied the equivalent scheduling policy to every process. Moreover, if the scheduling policy is once determined, it is unable to change without resetting the operating system which takes much time. In this paper, we propose an intelligent CPU process scheduling algorithm using fuzzy inference with user models. It classifies processes into three classes, batch, interactive and real-time processes, and models user’s preferences to each process class. Finally, it assigns the priority of each process according to the class of the process and user’s preference through the fuzzy inference. The experimental result shows the proposed method can adapt to user and allow different scheduling policies to multiple users.

Forecasting activities are frequent and widespread in our life. Since Song and Chissom proposed the fuzzy time series in 1993, many previous studies have proposed variant fuzzy time series models to deal with uncertain and vague data. A drawback of these models is that they do not consider appropriately the weights of fuzzy relations. This paper proposes a new method to build weighted fuzzy rules by computing cardinality of each fuzzy relation to solve above problems. The proposed method is able to build the weighted fuzzy rules based on concept of large itemsets of Apriori. The yearly data on enrollments at the University of Alabama are adopted to verify and evaluate the performance of the proposed method. The forecasting accuracies of the proposed method are better than other methods.

Fuzzy interpolative reasoning is an important research topic of sparse fuzzy rule-based systems. In recent years, some methods have been presented for dealing with fuzzy interpolative reasoning. However, the involving fuzzy sets appearing in the antecedents of fuzzy rules of the existing fuzzy interpolative reasoning methods must be normal and non-overlapping. Moreover, the reasoning conclusions of the existing fuzzy interpolative reasoning methods sometimes become abnormal fuzzy sets. In this paper, in order to overcome the drawbacks of the existing fuzzy interpolative reasoning methods, we present a new fuzzy interpolative reasoning method for sparse fuzzy rule-based systems. The proposed fuzzy interpolative reasoning method can handle the situation of non-normal and overlapping fuzzy sets appearing in the antecedents of fuzzy rules. It can overcome the drawbacks of the existing fuzzy interpolative reasoning methods in sparse fuzzy rule-based systems.

Support vector machine (SVM) is a very promising classification technique developed by Vapnik. However, there are still some shortcomings in the original SVM approach. First, SVM was originally designed for binary classification. How to extend it effectively for multi-class classification is still an on-going research issue. Second, SVM does not consider the distribution of each class. In this paper, we propose an extension to the SVM method of pattern recognition for solving the multi-class problem in one formal step. Contrast to previous multi-class SVMs, our approach considers the distribution of each class. Experimental results show that the proposed method is more suitable for practical use than other multi-class SVMs, especially for unbalanced datasets.

Ventilator weaning is the process of discontinuing mechanical ventilation from patients with respiratory failure. Previous investigations reported that 39%-40% of the intensive care unit (ICU) patients need mechanical ventilator for sustaining their lives. Among them, 90% of the patients can be weaned from the ventilator in several days while other 5%-15% of the patients need longer ventilator support. Modern mechanical ventilators are believed to be invaluable tools for stabilizing the condition of patients in respiratory failure. However, ventilator support should be withdrawn promptly when no longer necessary so as to reduce the likelihood of known nosocomial complications and costs. Although successful ventilator weaning of ICU patients has been widely studied, indicators for accurate prediction are still under investigation. Furthermore, the predication rate of successful weaning is only 35-60% based on previous studies. It is desirable to have objective measurements and predictors of weaning that decrease the dependence on the wisdom and skill of an individual physician. However, one study showed that clinicians were often wrong when predicting weaning outcome. In this study, 189 patients, who had been supported by mechanical ventilation for longer than 21 days and were clinically stable were recruited from our all-purpose ICUs. Twenty-seven variables in total were recorded, while only 8 variables which reached significant level were used for support vector machine (SVM) classification after logistic regression analysis. The result shows that the successful prediction rate achieves as high as 81.5% which outperforms a recently published predictor (78.6%) using combination of sample entropy of three variables, inspiratory tidal volume, expiratory tidal volume, and respiration rate.

Classification problem asks to construct a classifier from a given data set, where a classifier is required to capture the hidden oracle of the data space. Recently, we introduced a new class of classifiers ICF, which is based on iteratively composed features on {0,1, ∗ }-valued data sets. We proposed an algorithm ALG-ICF to construct an ICF classifier and showed its high performance. In this paper, we extend ICF so that it can also process real world data sets consisting of numerical and/or categorical attributes. For this purpose, we incorporate a discretization scheme into ALG-ICF as its preprocessor, by which an input real world data set is transformed into {0,1, ∗ }-valued one. Based on the experimental studies on conventional discretization schemes, we propose a new discretization scheme, integrated construction (IC). Our computational experiments reveal that the ALG-ICF equipped with IC outperforms a decision tree constructor C4.5 in many cases.

Chance discovery has achieved much success in discovering events that, though rare, are important to human decision making. Since humans are able to efficiently interact with graphical representations of data, it is useful to use visualizations for chance discovery. KeyGraph enables efficient visualization of data for chance discovery, but does not have provisions for adding domain-specific constraints. This contribution extends the concepts of KeyGraph to a visualization method based on the target sociogram. As the target sociogram is hierarchical in nature, it allows hierarchical constraints to be embedded in visualizations for chance discovery. The details of the hierarchical visualization method are presented and a class of problems is defined for its use. An example from software requirements engineering illustrates the efficacy of our approach.

Recently, mobile devices are regarded as a content storage with their functions such as camera, camcorder, and music player. It creates massive new data and downloads contents from desktop or wireless internet. Because of the massive size of digital contents in the mobile devices, user feels difficulty to recall or find information from the personal storage. If it is possible to organize the storage in a style of human-memory management, it could reduce user’s effort in contents management. Based on the evidence that human memory is organized as an episodic-style, we propose a KeyGraph-based reorganization method of mobile device storage for better accessibility to the data. It can help user not only find useful information from the storage but also expand his/her memory by adding user’s contexts such as location, SMS, call, and device status. User can recall his/her memory from the contents and contexts. KeyGraph finds rare but relevant events that can be used as a memory landmark in the episodic memory. Using artificially generated logs from a pre-defined scenario, the proposed method is tested and analyzed to check the possibility.

The activity of an organization is excited by introducing new persons. Understanding such catalyst personality is an important basis for fostering communication among groups with opposing preference. In the prior understanding, the groups are independent segments. Cognition, resulted from seeing the overlaps revealed between the segments, is not the same as the prior understanding. This gap is a clue. We demonstrate an experiment using questionnaire on the preference of art pieces.