Catálogo de publicaciones - libros

Based on the rough set theory, this paper introduces a multilayer rough-fuzzy rules design method to keep fuzzy rules dimension of every layer not more than three for consistency with man’s thinking characteristics, advantageous for understanding, checking and correcting rules. For rationally reducing and integrating input variables, the paper presents a rapid fuzzy rules extraction algorithm based on RST, to discover knowledge from sample database. This algorithm improves C-D indiscernible matrix. It introduces the computation program for core attributes. The program for quasi-optimal attribute reduction is presented, in which information increment of decision D is used as heuristic information of attributes selection to accelerate selective velocity of optimal attributes set. This multilayer fuzzy controller is combined with conventional PID, applied in unit control system of power plant. The simulation results show that the control system has higher control qualities with high speed, small overshoot and strong robustness.

The fuzzy Takagi-Sugeno-Kang model and the inference system proposed by these authors is a very powerful tool for function approximation problems due to its capability of expressing a complex nonlinear system using a set of simple linear rules. Nevertheless, during the learning and optimization process, usually a trade-off has to be carried out among global system accuracy and sub-models (rules) interpretability. In this paper we review the TaSe model [8] for function approximation (for Grid-Based Fuzzy Systems and extend it to consider Clustering-Based Fuzzy Systems) that is learned from an I/O numerical data set and that will allow us to extract strong interpretable rules, whose consequents are the Taylor Series Expansion of the model output around the rule centres. This TaSe model provides full interpretability to the local models with high accuracy in the global approximation. The rule extraction process using the TaSe model and its properties will be reviewed using a significant example.

In the field of cluster analysis, the fuzzy -means, -modes and -prototypes algorithms were designed for numerical, categorical and mixed data sets respectively. However, all the above algorithms assume that each feature of the samples plays an uniform contribution for cluster analysis. To consider the particular contributions of different features, a novel feature weighted fuzzy clustering algorithm is proposed in this paper, in which the ReliefF algorithm is used to assign the weights for every feature. By weighting the features of samples, the above three clustering algorithms can be unified, and better classification results can be also achieved. The experimental results with various real data sets illustrate the effectiveness of the proposed algorithm.

The work leading to this paper is semantic image classification. The aim is to evaluate contributions of clustering mechanisms to organize low-level features into semantically meaningful groups whose interpretation may relate to some description task pertaining to the image content. Cluster assignment reveals underlying structures in the data sets without requiring prior information. The semantic component indicates that some domain knowledge about the classification problem is available and can be used as part of the training procedures. Besides, data structural analysis can be applied to determine proximity and overlapping between classes, which leads to misclassification problems. This information is used to guide the algorithms towards a desired partition of the feature space and establish links between visual primitives and classes. It derives into partially supervised learning modes. Experimental studies are addressed to evaluate how unsupervised and partially supervised fuzzy clustering boost semantic-based classification capabilities.

A new algorithm of evolutionary computing, which combines clone selective algorithm involved in artificial immunity system theory and mind evolution algorithm (MEA) proposed in reference [4], is presented in this paper. Based on similartaxis which is the one of MEA operators, some operators borne by the new algorithm including clone mutation, clone crossover, clone selection, is also introduced. Then the clone mind evolution algorithm (CMEA) is developed by using the diversity principle of antigen-antibody. The simulating results of the representative evaluation function show that the problem of degeneration phenomenon existing in GA and MEA can be perfectly solved, and the rapidity of convergence is evidently improved by CMEA studied in the paper. In the example of the solution to the numerical problem, the search range of solution is expanded and the possibility of finding the optimal solution is increased.

The Estimation of Distribution Algorithm is a new class of population based search methods in that a probabilistic model of individuals are estimated based on the high quality individuals and used to generate the new individuals. In this paper we compute 1) some upper bounds on the number of iterations required for global convergence of EDA 2) the exact number of iterations needed for EDA to converge to global optima.

We propose a new algorithm to find minimal rough set reducts by using Particle Swarm Optimization (PSO). Like Genetic Algorithm, PSO is also a type of evolutionary algorithm. But compared with GA, PSO does not need complex operators as crossover and mutation that GA does, it requires only primitive and simple mathematical operators, and is computationally inexpensive in terms of both memory and times. The experiments on some UCI data compare our algorithm with GA-based, and other deterministic rough set reduction algorithms. The results show that PSO is efficient to minimal rough set reduction.

In this paper, Mind Evolutionary Algorithm (MEA) is introduced to correct the nonlinearity of voltage-controlled oscillator (VCO) in linear frequency modulation continuous wave (LFMCW) radar level gauge. Firstly, the frequency modulation (FM) voltage is divided into several subsections. By using fast Fourier transform (FFT) analysis for the beat frequency signals and distilling the characteristic of the spectrum, an evaluation function is constructed. Then MEA is applied to optimize the end coordinates of the subsections to achieve the nonlinear curve of FM voltage so as to compensate for the nonlinearity of VCO. Experiments show that the proposed method has good correction performance with no requirement of additional hardware and measuring equipment and is easy to apply.

This paper discusses the degree of granularity and dependence of contingency tables from the viewpoint of linear algebra. From the results of determinantal divisors, it seems that the devisors provide information on the degree of dependencies between the matrix of the whole elements and its submatrices and the increase of the degree of granularity may lead to that of dependence. However, this paper shows that a constraint on the sample size of a contingency table is very strong, which leads to the evaluation formula where the increase of degree of granularity gives the decrease of dependency.

We address the problem of determination of the size of the test set which can can guarantee statistically significant results in classifier error estimation and in selection of the best classifier from a given set. We focus on the case of the 0-1 valued loss function and we provide one and two sides optimal bounds for Validation (known also as Hold-Out Estimate and Train-and-Test Method). We also calculate the smallest sample size, necessary for obtaining the bound for given estimation accuracy and reliability of estimation, and we present the results in tables. Finally, we propose strategies for classifier design using the bounds derived.