Neuro-fuzzy approach to processing inputs with missing values in pattern recognition problems

An approach to dealing with missing data, both during the design and normal operation of a neuro-fuzzy classifier is presented in this paper. Missing values are processed within a general fuzzy min–max neural network architecture utilising hyperbox fuzzy sets as input data cluster prototypes. An emphasis is put on ways of quantifying the uncertainty which missing data might have caused. This takes a form of classification procedure whose primary objective is the reduction of a number of viable alternatives rather than attempting to produce one winning class without supporting evidence. If required, the ways of selecting the most probable class among the viable alternatives found during the primary classification step, which are based on utilising the data frequency information, are also proposed. The reliability of the classification and the completeness of information is communicated by producing upper and lower classification membership values similar in essence to plausibility and belief measures to be found in the theory of evidence or possibility and necessity values to be found in the fuzzy sets theory. Similarities and differences between the proposed method and various fuzzy, neuro-fuzzy and probabilistic algorithms are also discussed. A number of simulation results for well-known data sets are provided in order to illustrate the properties and performance of the proposed approach.     

Difference between canonical and grand canonical ensembles in discrete lattice gas models

We calculate the site occupation probabilities of one-dimensional lattice gas models within the canonical and grand canonical ensembles. The appearing differences do not vanish if we increase the system size keeping the site energies discrete. In this way one can explain the surprising numerical results of Barszczak and Kutner. This effect in the single-site occupation number disappears in higher dimensions.     

Bubble point pressure estimates from Gibbs ensemble simulations

Bubble pressure points of ethanol–1,1,1,2,3,3,3-heptafluoropropane (HFC-227ea refrigerant) mixtures from the third Industrial Fluid Properties Simulation Challenge are computed using publicly available molecular simulation software. Several published force fields are compared against the known answers provided in the contest guidelines and the best force fields are used to make predictions for the unknown results.     

偏最小二乘回归的贝叶斯正则化神经网络集成模型在证券分析预测中的应用

神经网络集成技术能有效地提高神经网络的预测精度和泛化能力,已经成为机器学习和神经计算领域的一个研究热点.利用Bagging技术和不同的神经网络算法生成集成个体,并用偏最小二乘回归方法从中提取集成因子,再利用贝叶斯正则化神经网络对其集成,以此建立上证指数预测模型.通过上证指数开、收盘价进行实例分析,计算结果表明该方法预测精度高、稳定性好.     

运用反例基于识别的手写数字串分割

采用基于识别的分割方法进行手写数字串分割.在识别的过程中,运用反例样本估计分类器参数,实验数据表明,这种运用反例样本训练的分类器与没有经过反例样本训练的分类器相比,将提高拒识率到19%左右,从而保证了较高的识别率,验证了只有经过反例训练的分类器的输出结果才是可信赖的.     

The Method of Virtual Experts in Mathematical Diagnostics

Mathematical Diagnostics (MD) deals with identification problems arising in different practical areas. Some of these problems can be described by mathematical models where it is required to identify points belonging to two or more sets of points. Most of the existing tools provide some identification rule (a classifier) by means of which a given point is assigned (attributed) to one of the given sets. Each classifier can be viewed as a virtual expert. If there exist several classifiers (experts), the problem of evaluation of experts’ conclusions arises. In the paper for the case of supervised classification the method of virtual experts (the VE-method) is described. Based on this method, a generalized VE method is proposed where each of the classifiers can be chosen from a given family of classifiers. As a result, a new optimization problem with a discontinuous functional is stated. Examples illustrating the proposed approach are provided. The work of the second author was supported by the Russian Foundation for Fundamental Studies (RFFI) under Grant No 03-01-00668.     

A complementary thermodynamic limit for classical Coulomb matter

The canonical equilibrium measure of classical two-component Coulomb matter with regularized interactions is analyzed in a finite volume. It is shown that, in the mean-field regime, the one-particle density is inhomogeneous on a new characteristic length scale _inh. For a system ofN positive andN negative particles, _inh and the characteristic length scale of correlations _corr (=Debye screening length) are related via _inh=(2N)^1/2 _corr. The major conceptual conclusion that is drawn from this is that one needs two nontrivial complementary thermodynamic limits to define the equilibrium thermodynamics of two-component Coulomb systems. One of them is the standard thermodynamic limit (infinite volume), where one takesN, _corr fixed. Its complementary limit is characterized byN, _inh fixed, and is a finite-volume inhomogeneous mean-field limit. The most prominent new feature in the mean-field thermodynamic limit, which is absent in the standard thermodynamic limit, is an anomalous first-order phase transition where the Coulomb system explodes or implodes, respectively. The phase transition is connected with the existence of a metastable plasma phase far below the ionization temperature.     

Surface properties of finite classical Coulomb systems: Debye-Hückel approximation and computer simulations

We report analytical and numerical studies of surface correlations in finite, homogeneously polarizable, classical Coulomb systems placed in an insulating or conducting environment. Their purpose is to understand the phenomenological, shape-dependent laws of electrostatics, from the point of view of statistical mechanics; we focus on the knowledge of the dielectric susceptibility of the system, a quantity proportional to the equilibrium fluctuation of the system's instantaneous polarization per unit volume. This goal has been achieved for a system in a conducting state. The picture is that the shape-dependent part of the susceptibilities results from the action of unbounded observables (the second moments of the instantaneous polarization of the system) on long-range surface correlations and that the relations of electrostatics are verified by means of shape-dependent thermodynamic limits. This picture is supported (i) by exact solutions and asymptotic analysis of the Debye-Hückel approximation of multicomponent plasmas in disks and spheres with insulating and conducting environment and also in ellipses in a vacuum, and (ii) by computer simulations of a one-component plasma in a disk with different environments, notably a conducting environment with permeable and impermeable wall. These observations have revealed for the first time the reason why the susceptibility of a conducting disk in a conductor with impermeable walls diverges linearly with the radius of the disk: this is due to the occurrence of long-range radial correlations in the conductor. These findings are quantitatively interpreted in terms of a novel canonical Debye-Huckel approximation as contrasted to the ordinary grand canonical version. Lastly a fresh look at the problem of the surface correlations of a conductor in a vacuum, which places the observer close to the surface of the conductor but in the vacuum, is presented and applied to the disk, the ellipse, the cylinder, the sphere, and the wedge.