Bayesian inference for hierarchical mixtures-of-experts with applications to regression and classification

This paper studies the problems of inference and prediction in a class of models known as hierarchical mixtures-of-experts (HME). The statistical model underlying an HME is a mixture model in which both the mixture coefficients and the mixture components are generalized linear models. Bayesian inference regarding an HME's parameters is presented in the contexts of regression and classification using Markov chain Monte Carlo methods. A benefit of this Bayesian approach is the ability to obtain a sample from the posterior distribution of any functional of the parameters of the given model. In this way, more information is obtained than provided by a point estimate. The methods are illustrated on a nonlinear regression problem and on a breast cancer classification problem. The results indicate that the HME showed good prediction performance, and also gave the additional benefit of providing for the opportunity to assess the degree of certainty of the model in its predictions.     

Perceptual organization of behavior: A hierarchical control model of coordinated action.

Compared the behavior of 4 college students and 2 faculty members (including the present author) with a hierarchical control system model of behavioral organization. Ss varied the position of 2 control handles simultaneously to keep the distance constant between 2 pairs of lines. Three variations on this basic experiment that illustrate some fundamental properties of coordinated action showed (1) how independent actions, compensating for unpredictable and undetectable disturbances, can produce a single behavioral result; (2) how the ability to produce a particular result is maintained when the connection between action and result is changed; and (3) how 2 independent outputs can appear to be related as coordinative structures when one output disturbs a result being controlled by the other. The correlation between the behavior of Ss and the model in all experiments was typically on the order of .99. A detailed examination of the operation of the model demonstrated that actions are structured by perception, not by central commands or equations of constraint. (35 ref) (PsycINFO Database Record (c) 2011 APA, all rights reserved)     

Evidence for a hierarchical structure underlying avoidance behavior.

In studies on avoidance learning, a warning signal is followed by an aversive unconditioned stimulus unless the participant performs a designated response. The authors examined whether avoidance behavior can be based on hierarchical knowledge, that is, knowledge about the conditions under which certain relations hold. In the present study, a single avoidance response had different effects depending on the nature of the warning signal. Results showed that participants acquired this hierarchical knowledge and used it to avoid negative outcomes. The results are in line with an occasion setting account of avoidance learning and can be explained also by a modified version of Lovibond's (2006) account of avoidance learning. (PsycINFO Database Record (c) 2010 APA, all rights reserved)     

A hierarchical factor analysis of foreman behavior.

"Ratings of [141] industrial foremen were made using a check list. Scores on 18 variables representing six hypothesized aspects of supervisory behavior were factor analyzed using the hierarchical factor model. A detailed explanation of the application of the hierarchical model is given. In terms of the present data, four factors were found, one a general or 'halo' factor and three interpretable factors. These were described in terms of their meaning and variance. The factors were designated as Social Relations, Technical Job Knowledge, and Administrative Skills and are similar to several previously reported. It is concluded that the hierarchical factor model is a useful technique for the analysis of intercorrelations of trait ratings." (PsycINFO Database Record (c) 2010 APA, all rights reserved)     

A hierarchical model of binocular rivalry

Binocular rivalry is the alternating percept that can result when the two eyes see different scenes. Recent psychophysical evidence supports the notion that some aspects of binocular rivalry bear functional similarities to other bistable percepts. We build a model based on the hypothesis (Logothetis & Schall, 1989; Leopold & Logothetis, 1996; Logothetis, Leopold & Sheinberg, 1996) that alternation can be generated by competition between top-down cortical explanations for the inputs, rather than by direct competition between the inputs. Recent neurophysiological evidence shows that some binocular neurons are modulated with the changing percept; others are not, even if they are selective between the stimuli presented to the eyes. We extend our model to a hierarchy to address these effects.     

Refining the construct of psychopathy: Towards a hierarchical model.

Psychopathy is characterized by diverse indicators. Clinical accounts have emphasized 3 distinct facets: interpersonal, affective, and behavioral. Research using the Psychopathy Checklist–Revised (PCL–R), however, has emphasized a 2-factor model. A review of the literature on the PCL–R and related measures of psychopathy, together with confirmatory factor analysis of PCL–R data from North American participants, indicates that the 2-factor model cannot be sustained. A 3-factor hierarchical model was developed in which a coherent superordinate factor, Psychopathy, is underpinned by 3 factors: Arrogant and Deceitful Interpersonal Style, Deficient Affective Experience, and Impulsive and Irresponsible Behavioral Style. The model was cross-validated on North American and Scottish PCL–R data, Psychopathy Screening Version data, and data derived from the Diagnostic and Statistical Manual of Mental Disorders (4th ed.; American Psychiatric Association, 1994) antisocial personality disorder field trial. (PsycINFO Database Record (c) 2011 APA, all rights reserved)     

Non-schmid effects on the behavior of polycrystals—with applications to ni3al

The elastoviscoplastic single crystal constitutive model incorporatingnon-Schmid effects developed by Dao and Asaro (Mater. Sci. Eng. A, 1993, vol. 170, pp. 143-60) is introduced into Asaro and Needleman’s (Acta Metall., 1985, vol. 33, pp. 923-53) Taylor-like polycrystal model as well as Harren and Asaro’s (J. Mech. Phys. Solids, 1989, vol. 37, pp. 191-232) finite element polycrystal model. The single crystal non-Schmid effects, strain hardening, latent hardening, and rate sensitivity, are all described on the individual slip system level, while polycrystal mechanical properties on macroscale are predicted. In general, it is found that non-Schmid effects can have important influences on the “constant offset plastic strain yield surfaces,” stress-strain behavior, texture development, and shear band formation. Finite element calculations show that with moderate non-Schmid effects, localized deformation within a polycrystal aggregate tends to initiate earlier and form sharper and more intense shear bands. Heavy shear banding is found to produce less pronounced textures, which is consistent with existing experimental evidence on Ni₃Al. Examples with Ni₃Al demonstrate that the kind of non-Schmid effects existing in Ni₃Al can increase the generalized Taylor factor to values much higher than 3.06, raise the polycrystal strain hardening rate much higher than that which would be obtained using Schmid’s rule, and influence the deformation texture.     

A truncated poisson regression model with applications to occurrence of adenomatous polyps

Adenomatous polyps are considered as the dominant precursor lesion of colorectal cancer. A phase III colorectal cancer prevention trial, conducted by the Arizona Cancer Center, concerns the ability of wheat bran fibre supplement to reduce the recurrence of adenomatous polyps. All participants in the study are to have had colorectal polyps detected and removed during a baseline (qualifying) colonoscopy within three months prior to enrolment. In this paper, our interest focuses on occurrence of adenomatous polyps at the baseline colonoscopy. We use a truncated Poisson model to fit these types of data. We develop a regression model to assess the effects of explanatory factors on the positive counting variable. We fit truncated Poisson parameters by a log-linear regression model and estimate regression parameters by the maximum likelihood procedure. Finally, we apply it to the baseline colonoscopy data from the Wheat Bran Fiber study.     

Experimental research on the seismic behavior of CSPSWs connected to frame beams

The seismic performance of composite steel plate shear walls (CSPSWs) that consist of a steel plate shear wall (SPSW) with reinforced concrete (RC) panels attached to one or both sides by means of bolts or connectors is experimentally studied. The shear wall is connected to the frame beams but not to the columns. This arrangement restrains the possible outof-plane buckling of the thin-walled steel plate, thus significantly increasing the bearing capacity and ductility of the overall wall, and prevents the premature overall or local buckling failure of the frame columns. From a practical viewpoint, these solutions can provide open space in a floor as this type of composite shear walls with a relatively small aspect ratio can be placed parallel along a bay. In this study, four CSPSWs and one SPSW were tested and the results showed that both CSPSWs and SPSW possessed good ductility. For SPSW alone, the buckling appeared and resulted in a decrease of bearing capacity and energy dissipation capacity. In addition, welding stiffeners at comers were shown to be an effective way to increase the energy dissipation capacity of CSPSWs.     

基于二叉树型分层BP模型的板形模式识别

针对传统最小二乘多项式板形模式识别方法鲁棒性差、各分项物理意义不明确,以及普通BP(back propagation)识别法精度低等问题,选用勒让德多项式作为板形基本模式,提出一种基于二叉树型分层BP的板形模式识别并行计算模型.该模型通过逐层细化预测范围并选用多个神经网络进行递推.实验结果表明,采用此方法不仅增强了系统的抗干扰能力,而且提高了系统的识别精度.