Assessing and improving the stability of chemometric models in small sample size situations

Small sample sizes are very common in multivariate analysis. Sample sizes of 10–100 statistically independent objects (rejects from processes or loading dock analysis, or patients with a rare disease), each with hundreds of data points, cause unstable models with poor predictive quality. Model stability is assessed by comparing models that were built using slightly varying training data. Iterated k-fold cross-validation is used for this purpose. Aggregation stabilizes models. It is possible to assess the quality of the aggregated model without calculating further models. The validation and aggregation methods investigated in this study apply to regression as well as to classification. These techniques are useful for analyzing data with large numbers of variates, e.g., any spectral data like FT-IR, Raman, UV/VIS, fluorescence, AAS, and MS. FT-IR images of tumor tissue were used in this study. Some tissue types occur frequently, while some are very rare. They are classified using LDA. Initial models were severely unstable. Aggregation stabilizes the predictions. The hit rate increased from 67% to 82%.     

A spectral similarity measure using Bayesian statistics

A spectral similarity measure was developed that can differentiate subtle differences between two spectra. The spectra are digitalized into a vector. The difference between the two spectra is defined by a difference vector, which is one spectrum minus the other. The spectral similarity measure is transformed into a hypothesis test of the similarities and differences between the two spectra. The scalar mean of the difference vector is used as the statistical variable for the hypothesis test. A threshold for the hypothesis that the spectra are different was proposed. The Bayesian prior odds ratio was estimated from multiple spectra of the same sample. The posterior odds ratio was used to quantity the spectral similarity measure of the two spectra. Diffuse reflectance near-infrared spectra of tobacco samples of two formulations were used to demonstrate this method. The results show that this new method can detect subtle differences between the spectra.     

广义Gelfand模型的正解

在不要求极限limτ→0f(τ）/τ,limτ→∞f（τ）τ存在的情况下,讨论了二阶边值问题u″(t) λh(t)f（u（t））=0,0≤t≤1,u（0）=u（1）=0的正解存在性和多重性。     

Statistical aspects of the identification of material parameters for elasto-plastic models

Summary The presented method to identify material parameters for inelastic deformation laws is based on the numerical analysis of inhomogeneous stress and strain fields received from suitable experiments. Tensile and bending tests were carried out to obtain elastic and hardening parameters. The deformation law for small elasto-plastic strains is presented as a system of nonlinear differential and algebraic equations (DAE) consisting of the stress–strain relation, evolution equations for the internal variables and the yield condition. Different rules for the evolution equations of isotropic, kinematic and distorsional hardening are proposed. The DAE are discretized using an implicit Euler method, and the resulting system of nonlinear algebraic equations is solved using the Newton method. Deterministic optimization procedures are preferred to identify material parameters from a least-squares functional of numerical and measured comparative quantities. The gradient of the objective function was calculated using a semianalytical sensitivity analysis. Due to measurement errors, the optimal sets of material parameters are non unique. The approximate estimation of confidence regions and the calculation of correlation coefficients is presented. The results of several optimization processes for material parameters of elasto-plastic deformation laws show a good agreement between measured and calculated values, but they show also problems which may occur if systematic errors will not be recognized and deleted. Received 30 September 1999; accepted for publication 8 March 2000     

Spectral distributions in nuclei: General principles and applications

The subject of spectral distribution methods where one derives and applies the locally smoothed forms of observables in nuclei is briefly reviewed. It is well understood that the local forms (with respect to energy) of the level density function, expectation values and strength densities are Gaussian, linear (or ratio of Gaussians) and a bivariate Gaussian respectively. To accomodate symmetries in the above forms, one has to deal with multivariate distributions in general; for example the angular-momentum (J) decomposition leads to a bivariate Gaussian form for the level density. These results extend to indefinitely large spaces by method of partitioning and they generate convolution forms. The origin of these remarkable spectral properties is discussed and shell model examples are given to substantiate their applicability to nuclear systems. Spectral distribution theory is a practical, usable theory because the smoothed forms are defined in terms of traces of low particle-rank operators, and the trace information propagates. Finally we discuss the application of the spectral methods for a wide range of nuclear problems; these include binding energies, orbit occupancies, electromagnetic andβ-decay sum rule quantities, analysis of operators, symmetry breaking, numerical level densities, and determination of bounds on time-reversal non-invariant part of nucleon-nucleon interaction.     

Lattice gas generalization of the hard hexagon model. II. The local densities as elliptic functions

In a previous paper we considered an extension of the hard hexagon model to a solvable two-dimensional lattice gas with at most two particles per pair of adjacent sites. Here we use various mathematical identities (in particular Gordon's generalization of the Rogers-Ramanujan relations) to express the local densities in terms of elliptic functions. The critical behavior is then readily obtained.Supported in part by the National Science Foundation Grant MCS 8201733.     

Minimum SNR and acquisition for bias-free estimation of fractional anisotropy in diffusion tensor imaging - a comparison of two analytical techniques and field strengths

Although it is known that low signal-to-noise ratio (SNR) can affect tensor metrics, few studies reporting disease or treatment effects on fractional anisotropy (FA) report SNR; the implicit assumption is that SNR is adequate. However, the level at which low SNR causes bias in FA may vary with tissue FA, field strength and analytical methodology. We determined the SNR thresholds at 1.5 T vs. 3 T in regions of white matter (WM) with different FA and compared FA derived using manual region-of-interest (ROI) analysis to tract-based spatial statistics (TBSS), an operator-independent whole-brain analysis tool. Using ROI analysis, SNR thresholds on our hardware-software magnetic resonance platforms were 25 at 1.5 T and 20 at 3 T in the callosal genu (CG), 40 at 1.5 and 3 T in the anterior corona radiata (ACR), and 50 at 1.5 T and 70 at 3 T in the putamen (PUT). Using TBSS, SNR thresholds were 20 at 1.5 T and 3 T in the CG, and 35 at 1.5 T and 40 at 3 T in the ACR. Below these thresholds, the mean FA increased logarithmically, and the standard deviations widened. Achieving bias-free SNR in the PUT required at least nine acquisitions at 1.5 T and six acquisitions at 3 T. In the CG and ACR, bias-free SNR was achieved with at least three acquisitions at 1.5 T and one acquisition at 3 T. Using diffusion tensor imaging (DTI) to study regions of low FA, e.g., basal ganglia, cerebral cortex, and WM in the abnormal brain, SNR should be documented. SNR thresholds below which FA is biased varied with the analytical technique, inherent tissue FA and field strength. Studies using DTI to study WM injury should document that bias-free SNR has been achieved in the region of the brain being studied as part of quality control.     

基于前失速先兆检测机理的叶顶喷气扩稳实验研究

拓宽轴流压气机稳定工作裕度及探索其流动失稳控制途径一直是内部流动研究的焦点。本文从前失速先兆检测机理入手,对前失速先兆与叶顶间隙泄漏流非定常性进行了关联性研究,发现了前失速先兆与叶顶间隙泄漏流非定常性有一定的关联性。在认识这类关联性的基础上,提出了基于前失速先兆捕捉的失稳控制途径,并结合DSP控制器构建扩稳控制方案,在低速轴流压气机上进行了实验验证。实验结果表明:采用叶顶喷气扩稳控制措施后,简化了原来复杂的控制机构,并为控制措施的实施预留了反馈控制时间,与原来的定常喷气控制措施相比有变工况自适应的优势。     

Free energy of the solvable chiral Potts model

Very recently, it has been shown that there are chiralN-state Potts models in statistical mechanics that satisfy the star-triangle relation. Here it is shown that the relation implies that the free energy (and its derivatives) satisfies certain functional relations. These can be used to obtain the free energy: in particular, we expand about the critical case and find that the exponent is 1–2/N.