Search

In this paper, we compare some traditional statistical methods for predicting financial distress to some more “unconventional” methods, such as decision tree classification, neural networks, and evolutionary computation techniques, using data collected from 200 Taiwan Stock Exchange Corporation (TSEC) listed companies. Empirical experiments were conducted using a total of 42 ratios including 33 financial, 8 non-financial and 1 combined macroeconomic index, using principle component analysis (PCA) to extract suitable variables.This paper makes four critical contributions: (1) with nearly 80% fewer financial ratios by the PCA method, the prediction performance is still able to provide highly-accurate forecasts of financial bankruptcy; (2) we show that traditional statistical methods are better able to handle large datasets without sacrificing prediction performance, while intelligent techniques achieve better performance with smaller datasets and would be adversely affected by huge datasets; (3) empirical results show that C5.0 and CART provide the best prediction performance for imminent bankruptcies; and (4) Support Vector Machines (SVMs) with evolutionary computation provide a good balance of high-accuracy short- and long-term performance predictions for healthy and distressed firms. Therefore, the experimental results show that the Particle Swarm Optimization (PSO) integrated with SVM (PSO–SVM) approach could be considered for predicting potential financial distress.

Uniform resampling is the easiest to apply and is a general recipe for all problems, but it may require a large replication size B. To save computational effort in uniform resampling, balanced bootstrap resampling is proposed to change the bootstrap resampling plan. This resampling plan is effective for approximating the center of the bootstrap distribution. Therefore, this paper applies it to neural model selection. Numerical experiments indicate that it is possible to considerably reduce the replication size B. Moreover, the efficiency of balanced bootstrap resampling is also discussed in this paper.

The purpose of this paper is to propose a new method for solving multiple-attribute decision making (MADM) problems with trapezoid fuzzy linguistic variables. To begin with, this paper reviews the concepts of the trapezoid fuzzy linguistic variables and the possibility degree matrix. Then, some operators and their characteristics are introduced, for aggregating the trapezoid fuzzy linguistic variables, such as the trapezoid fuzzy linguistic averaging (TFLA) operator, the trapezoid fuzzy linguistic weighted averaging (TFLWA) operator, and the trapezoid fuzzy linguistic ordered weighted averaging (TFLOWA) operator. On the basis of the disadvantages of these operators, the trapezoid fuzzy linguistic hybrid ordered weighted averaging (TFLHOWA) operator is proposed. Furthermore, a new method for solving MADM problems with the trapezoid fuzzy linguistic variables is proposed, based on the TFLHOWA operator. Finally, an illustrative example is given, to verify the feasibility and effectiveness of the method developed.

In this article, we will show that (1) the results about the triple I method for fuzzy reasoning obtained in the paper “Computers and Mathematics with Applications 44 (2002) 1567–1579” are correct; (2) Example 2.1 in the paper “Notes on triple I method of fuzzy reasoning” presented by Hua-wen Liu is incorrect; (3) α(u,v)-triple I method presented by Liu is a generalization of α-triple I method.

In this paper, a novel approach to adjusting the weightings of fuzzy neural networks using a Real-coded Chaotic Quantum-inspired genetic Algorithm (RCQGA) is proposed. Fuzzy neural networks are traditionally trained by using gradient-based methods, which may fall into local minimum during the learning process. To overcome the problems encountered by the conventional learning methods, RCQGA algorithms are adopted because of their capabilities of directed random search for global optimization. It is well known, however, that the searching speed of the conventional quantum genetic algorithms (QGA) is not satisfactory. In this paper, a real-coded chaotic quantum-inspired genetic algorithm (RCQGA) is proposed based on the chaotic and coherent characters of Q-bits. In this algorithm, real chromosomes are inversely mapped to Q-bits in the solution space. Q-bits probability-guided real cross and chaos mutation are applied to the evolution and searching of real chromosomes. Chromosomes consisting of the weightings of the fuzzy neural network are coded as an adjustable vector with real number components that are searched by the RCQGA. Simulation results have shown that faster convergence of the evolution process in searching for an optimal fuzzy neural network can be achieved. Examples of nonlinear functions approximated by using the fuzzy neural network via the RCQGA are demonstrated to illustrate the effectiveness of the proposed method.

This article has been retracted: please see Elsevier Policy on Article Withdrawal (http://www.elsevier.com/locate/withdrawalpolicy).This article has been retracted at the request of the Editor-in-Chief.The authors have plagiarized part of a paper that had already appeared in J. Comput. Appl. Math., 228 (2009) 226–230 doi:10.1016/j.cam.2008.09.015). One of the conditions of submission of a paper for publication is that authors declare explicitly that their work is original and has not appeared in a publication elsewhere. Re-use of any data should be appropriately cited. As such this article represents a severe abuse of the scientific publishing system. The scientific community takes a very strong view on this matter and apologies are offered to readers of the journal that this was not detected during the submission process.

This paper reflects results of research into the construction and analysis of 〈X, R〉 models within the framework of a general approach to solving optimization problems with fuzzy coefficients. This approach involves a modification of traditional mathematical programming methods and is associated with formulating and solving one and the same problem within the framework of mutually interrelated models. The use of the approach allows one to maximally cut off dominated alternatives. The subsequent contraction of the decision uncertainty region is based on reducing the problem to models of multiobjective choosing alternatives in a fuzzy environment with the use of fúzzy preference relation techniques for analyzing these models. Three techniques for fuzzy preference modeling are discussed in the paper. The first technique is based on the construction of membership functions of subsets of nondominated alternatives with simultaneous considering of all criteria (fuzzy preference relations). The second technique is of a lexicographic character and consists of step-by-step introducing of fuzzy preference relations. The third technique is based on aggregating membership functions of subsets of nondominated alternatives corresponding to each preference relation. These techniques have served for developing a corresponding system for multiobjective decision making (MDMS). C++ windows of the MDMS are presented for input, output, and some intermediate procedures. The results of the paper are of a universal character and are already being used to solve power engineering, naval engineering, and management problems.

This paper presents a survey of feature saliency measures used in artificial neural networks. Saliency measures can be used for assessing a feature's relative importance. In this paper, we contrast two basic philosophies for measuring feature saliency or importance within a feed-forward neural network. One philosophy is to evaluate each feature with respect to relative changes in either the neural network's output or the neural network's probability of error. We refer to this as a derivative-based philosophy of feature saliency. Using the derivative-based philosophy, we propose a new and more efficient probability of error measure. A second philosophy is to measure the relative size of the weight vector emanating from each feature. We refer to this as a weight-based philosophy of feature saliency. We derive several unifying relationships which exist within the derivative-based feature saliency measures, as well as between the derivative and the weight-based feature saliency measures. We also report experimental results for an target recognition problem using a number of derivative-based and weight-based saliency measures.

In this paper, we describe new results of research on geometrical properties of complex control systems, the so-called Linguistic Geometry. This research includes the development of syntactic tools for knowledge representation and reasoning about large-scale hierarchical complex systems. It relies on the formalization of search heuristics of high-skilled human experts that have resulted in the development of successful applications in different areas. A hierarchy of subsystems of a complex system, the networks of paths, is represented as a hierarchy of formal languages. In this paper, we investigate transformations of these networks while a system moves from one state to another. The investigation consists of formal, constructive separation of changed and unchanged parts of system representation, the hierarchy of languages. Thus, we address a problem relative to the well-known Frame Problem for planning systems. A partial solution is presented in the form of the theorem about translations of network languages. Formal considerations are illustrated by example of Air Force robotic vehicles.

The objective of the research considered in this paper is to develop a theoretical foundation for the representation of large-scale hierarchical complex systems, the so-called Linguistic Geometry. The research relies on the formalization of heuristics of highly-skilled human experts which have resulted in the development of successful decision support systems. This approach is based on a broad application of the theory of formal languages and grammars as well as theories of formal problem-solving and planning on the basis of the first-order predicate calculus. This paper reports new results in the investigation of geometrical properties of the first-level subsystems (paths of elements) unified as one-dimensional Linguistic Geometry. Theoretical constructions considered in this paper are illustrated employing comprehensive examples from power maintenance scheduling, robot control, and the chess game. A program implementation of this approach should generate decision support systems for a wide class of practical problems.

Four methods for the associative retrieval of semantic networks are described. These methods differ from those traditional approaches, such as SNEPS, in which an entire knowledge base is treated as a single network. Here the knowledge base is viewed as an organized collection of networks and is most appropriate for applications (such as bibliographic retrieval) in which pieces of knowledge need to be treated individually. Method I is an arbitrary flat ordering of database graphs, Method II a two-level ordering, and Method III is a full partial order. Method IV is a novel method known as ``hierarchical node descriptor method'''' that is based on the ``refinement'''' method of subgraph-isomorphism. A ``pattern associativity'''' principle explains the development and effectiveness of each of these methods. Moving from Method I through Method IV there is a steady increase in both pattern associativity and efficiency. A theorem is proven that establishes the superiority of Method III over Method II despite the fact that Method II is the method most often used. A brief discussion of how parallelism may be incorporated also accompanies the description of each method. Most of the paper applies these methods to conceptual graphs and a later section shows how the techniques can be extended to other semantic-network formalisms. The paper concludes by showing how generalization graphs constructed through pattern associativity may also have semantic validity in the domains from which they have been derived. (Supersedes 90-30.)

This paper presents an overview of the ECO (English COnversational System) family formalism of semantic network. In the paper, we describe the components of our semantic network, discussing its suitability as a representation of propositional knowledge. The use of our semantic network as a uniform representation mediating between specialised representations appropriate to particular task domains (e.g., understanding natural languages, etc.) is discussed. We motivate and explain a comprehensive network formalism. Special problems with respect to the use of logical connectives, quantifiers, descriptions, modalities, and certain other constructions that fail in conventional semantic networks, are systematically resolved with extensions to conventional network notations. The representation harmonizes with linear one-dimensional logical notations, illustrating the close kinship of the two notations. This kinship supports the claim that networks have inherited formal interpretability from logical notations. Several issues of network form and content, which are more fundamental than the choice of a network syntax, are addressed. These issues are: (i) primitive versus nonprimitive representations; (ii) the separation of propositional content of text from pragmatic aspects; and (iii) network normal form versus ad hoc systems. The design of computer systems for specific tasks depends in part on early commitments to these issues. The succinctness, clarity, and intuitive nature of semantic networks argues in their favour if only for purely methodological advantages. Semantic networks are readable; they suggest procedures for comprehension and inference, and the computer data structures which they resemble. Examples will demonstrate how associative processing algorithms and complex pattern matching operations are readily identifiable using networks. These examples are given in the context of natural language understanding utilizing networks in a state-based conceptual representation. We discuss how to superimpose organisational strategies into the network representations, beginning with the representation of lexical information and extending to the superimposition of topical organisations in the knowledge base. Several special purpose inference mechanisms extend the topical organisation we superimpose on concepts to aid retrieval of other types of information about concepts. The use of networks is assessed and promising areas for future research are described.

In this paper, we consider a coarse grid approximation (numerical homogenization and multiscale finite element method) for the poroelasticity problem with stochastic properties. The proposed method is based on the construction of deep neural network for fast calculation of macroscopic parameters for a coarse grid approximation of the problem. We train neural networks on a set of selected realizations of local microscale stochastic fields and macroscale characteristics (effective property tensor or local matrix). We construct a deep learning method through a convolutional neural network (CNN) to learn a map between stochastic fields and macroscopic parameters. Numerical results are presented for two and three-dimensional model problems and show that the proposed method provides fast and accurate effective property predictions.

Many optimization procedures presume the availability of an initial approximation in the neighborhood of a local or global optimum. Unfortunately, finding a set of good starting conditions is itself a nontrivial proposition. Our previous papers [1,2] describe procedures that use simple and recurrent associative memories to identify approximately solutions to closely related linear programs. In this paper, we compare the performance of a recurrent associative memory to that of a feed-forward neural network trained with the same data. The neural network's performance is much less promising than that of the associative memory. Modest infeasibilities exist in the estimated solutions provided by the associative memory, but the basic variables defining the optimal solutions to the linear programs are readily apparent.

This paper discusses learning in the context of a diagnostic expert system. The diagnostic expert system is an example of a generate-and-test problem solver. Fault diagnosis heuristics (i.e. logical implications representing association between unusual features and failed components) hypothesize potential faults. The potential faults are verified or denied by comparing the predictions of a qualitative simulation to observe data. Learning in this context consists of modifying the fault diagnosis heuristics. This paper describes how heuristic rules and device models can be represented and revised in a logic programming framework. In addition, we demonstrate how logic programming can be extended to perform abductive reasoning in addition to deductive reasoning. Finally, we compare failure-driven learning and learning from successes for acquiring fault diagnosis heuristics via explanation-based learning.

It is still a challenging task to segment ultrasound medical images automatically and accurately because of the poor quality of images. To address these problems, a two-stage automatic segmentation scheme based on local entropy and a proposed active contour model for ultrasound images are put forward in this paper. First, a new region-based active contour model in the level set formulation, driven by global and local intensity information, is established for the segmentation. Furthermore, for automatically segmenting the ultrasound images, a coarse segmentation is performed utilizing local entropy information of the ultrasound images, then the coarse segmentation result is used as the initial value of the explicit segmentation based on the proposed model. Several experiments on real ultrasound images demonstrate that the proposed method outperforms other methods on both visual perception and objective evaluation metrics.

Color transfer methods usually suffer from spatial color coherency problem. In order to address this problem, this paper develops a fused color transfer method for image colorization. Our idea is to design a local student’s t-test to screen the incoherent colors in the preliminary colorization results obtained by a simple color transfer method with DFT and variance features. Furthermore, we propose a variational fusion model to inpaint these incoherent colors and fuse the other useful colors together. We also present an efficient algorithm for solving the fusion model numerically, and show the convergence of the algorithm. Finally, experimental results are reported to demonstrate the effectiveness of the proposed method, and its performance is competitive with those of the other testing methods.

The restoration of images degraded by blur and multiplicative noise is a critical preprocessing step in medical ultrasound images which exhibit clinical diagnostic features of interest. This paper proposes a novel non-smooth non-convex variational model for ultrasound images denoising and deblurring motivated by the successes of sparse representation of images and FoE based approaches. Dictionaries are well adapted to textures and extended to arbitrary image sizes by defining a global image prior, while FoE image prior explicitly characterizes the statistics properties of natural image. Following these ideas, the new model is composed of the data-fidelity term, the sparse and redundant representations via learned dictionaries, and the FoE image prior model. The iPiano algorithm can efficiently deal with this optimization problem. The new proposed model is applied to several simulated images and real ultrasound images. The experimental results of denoising and deblurring show that proposed method gives a better visual effect by efficiently removing noise and preserving details well compared with two state-of-the-art methods.

Due to the abundant noise, blurry boundaries, and intensity inhomogeneities present in ultrasound (US) images, it is a difficult task to segment US images accurately. In this paper, we propose a novel active contour method that combines global and local region information to achieve this task. The global information can segment US images with noise and blurry boundaries; while the local information can settle the intensity homogeneities of images. The proposed method can be directly applied to synthetic, real, and US-images segmentation. Results demonstrate the superiority of the proposed method over other representative algorithms. Moreover, we also extend the proposed method to vector-valued images. Experiments are performed to testify the feasibility of the method, and the proposed vector-valued idea can be applied to the medical co-segmentation in the future.

This paper presents a new level set formulation for active contour models (ACM). We propose the idea of integrating the eigenvalue information of Hessian matrix into the level set function. By this new level set function, the principal curvature information of images is used to enhance the ability of segmenting boundary regions. The advantages of our model are as follows: firstly, the interior and exterior object boundaries can be segmented with the initial contour being anywhere in the input image. Secondly, this method can work with heterogeneous images. Thirdly, the proposed model can produce smooth and right boundaries of objects having vital importance in medical operations. Extensive experiments demonstrate that the proposed model can obtain better segmentation results.

Most of local region-based active contour models in terms of the level set approach are able to segment images with intensity inhomogeneity. However, these models do not utilize global statistical information and are quite sensitive to the initial placement of the contour. This paper presents a new global and local region-based active contour model to segment images with intensity inhomogeneity. First, in order to reduce number of iterations, the global energy functional of the ChanVese (CV) model is used as the global term. Then, the local term is proposed to incorporate both local spatial information and local intensity information to handle intensity inhomogeneity. Moreover, to increase robustness of the initialization of the contour and reduce number of iterations, a convex energy functional and the dual algorithm are designed in the numerical implementation. Experimental results for synthetic and medical images have shown the efficiency and robustness of the proposed method.

In this paper, we present a new approach of multi-frame super-resolution (SR). The SR techniques strongly depend on the availability of accurate motion estimation. When the estimation of motion is not well established, as usually happens for non-parametric motion, annoying artifacts appear in the super-resolved image. Since SR problems suffer from the motion and blur estimations, new techniques are considered to improve the registration and restoration steps. The proposed method consists of a non-parametric image registration based on diffusion regularization and a nonlocal Laplace regularizer combined with a bilateral filter (BTV) in the reconstruction step to remove noise and motion outliers. The diffusion registration is employed to handle the small deformation between the unregistered images, while the combination of nonlocal Laplace and BTV is used to increase the robustness of the restoration step with respect to the blurring effect and to the noise. We also prove the existence of a solution to the well posed registration problem. Simulation results using different images show the effectiveness and robustness of our algorithm against noise and outliers compared to other existing methods.

The generic term fuzzy quantifier is employed in this paper to denote the collection of quantifiers in natural languages whose representative elements are: several, most, much, not many, very many, not very many, few, quite a few, large number, small number, close to five, approximately ten, frequently, etc. In our approach, such quantifiers are treated as fuzzy numbers which may be manipulated through the use of fuzzy arithmetic and, more generally, fuzzy logic.A concept which plays an essential role in the treatment of fuzzy quantifiers is that of the cardinality of a fuzzy set. Through the use of this concept, the meaning of a proposition containing one or more fuzzy quantifiers may be represented as a system of elastic constraints whose domain is a collection of fuzzy relations in a relational database. This representation, then, provides a basis for inference from premises which contain fuzzy quantifiers. For example, from the propositions “Most U's are A's” and “Most A's are B's,” it follows that “Most2 U's are B's,” where most2 is the fuzzy product of the fuzzy proportion most with itself.The computational approach to fuzzy quantifiers which is described in this paper may be viewed as a derivative of fuzzy logic and test-score semantics. In this semantics, the meaning of a semantic entity is represented as a procedure which tests, scores and aggregates the elastic constraints which are induced by the entity in question.