基于正则代数重建算法的二维温度场重建

在使用代数重建算法(Algebraic Reconstruction Technique,ART)对二维非均匀温度场进行重建时,离散误差和投影噪声会随着迭代修正被引入,为了平衡离散误差,减小算法对噪声的敏感度,在ART中引入了正则化项,并使用留一交叉验证法对单位正则化参数进行了选取,根据投影穿过待测区域路径的长度和单位正则化参数动态调整每条投影的正则化权重,实现了对每条投影离散误差和噪声水平的衡量。在不同的投影分布情况下,使用该算法对高斯单峰对称和高斯单峰偏置温度场进行了仿真重建,重建结果表明相比于传统迭代算法,该算法可有效提高温度场的重建精度,并且具有较好的稳定性。     

Leave-one-out manifold regularization

The manifold regularization (MR) based semi-supervised learning could explore structural relationships from both labeled and unlabeled data. However, the model selection of MR seriously affects its predictive performance due to the inherent additional geometry regularizer of labeled and unlabeled data. In this paper, two continuous and two inherent discrete hyperparameters are selected as optimization variables, and a leave-one-out cross-validation (LOOCV) based Predicted REsidual Sum of Squares (PRESS) criterion is first presented for model selection of MR to choose appropriate regularization coefficients and kernel parameters. Considering the inherent discontinuity of the two hyperparameters, the minimization process is implemented by using a improved Nelder-Mead simplex algorithm to solve the inherent discrete and continues hybrid variables set. The manifold regularization and model selection algorithm are applied to six synthetic and real-life benchmark dataset. The proposed approach, leveraged by effectively exploiting the embedded intrinsic geometric manifolds and unbiased LOOCV estimation, outperforms the original MR and supervised learning approaches in the empirical study.     

An experimental comparison of cross-validation techniques for estimating the area under the ROC curve

Reliable estimation of the classification performance of inferred predictive models is difficult when working with small data sets. Cross-validation is in this case a typical strategy for estimating the performance. However, many standard approaches to cross-validation suffer from extensive bias or variance when the area under the ROC curve (AUC) is used as the performance measure. This issue is explored through an extensive simulation study. Leave-pair-out cross-validation is proposed for conditional AUC-estimation, as it is almost unbiased, and its deviation variance is as low as that of the best alternative approaches. When using regularized least-squares based learners, efficient algorithms exist for calculating the leave-pair-out cross-validation estimate.     

分类器准确率评估的研究

由于机器学习中过度拟合问题的广泛存在,使得对分类器进行准确率评估显得十分重要.在对现有的两种评估方法-"k-fold交叉验证"和"随机子抽样"进行研究之后,设计出一种新的评估方法,旨在综合已有两种方法的优点,寻找一种更合理的评估方法,并以决策树构造的分类器进行验证.实验结果表明,在训练样本不多的情况下,新的评估方法应用...     

Linearly Combining Density Estimators via Stacking

This paper presents experimental results with both real and artificial data combining unsupervised learning algorithms using stacking. Specifically, stacking is used to form a linear combination of finite mixture model and kernel density estimators for non-parametric multivariate density estimation. The method outperforms other strategies such as choosing the single best model based on cross-validation, combining with uniform weights, and even using the single best model chosen by Cheating and examining the test set. We also investigate (1) how the utility of stacking changes when one of the models being combined is the model that generated the data, (2) how the stacking coefficients of the models compare to the relative frequencies with which cross-validation chooses among the models, (3) visualization of combined effective kernels, and (4) the sensitivity of stacking to overfitting as model complexity increases.     

GAKREM: A novel hybrid clustering algorithm

We introduce a novel clustering algorithm named GAKREM (Genetic Algorithm K-means Logarithmic Regression Expectation Maximization) that combines the best characteristics of the K-means and EM algorithms but avoids their weaknesses such as the need to specify a priori the number of clusters, termination in local optima, and lengthy computations. To achieve these goals, genetic algorithms for estimating parameters and initializing starting points for the EM are used first. Second, the log-likelihood of each configuration of parameters and the number of clusters resulting from the EM is used as the fitness value for each chromosome in the population. The novelty of GAKREM is that in each evolving generation it efficiently approximates the log-likelihood for each chromosome using logarithmic regression instead of running the conventional EM algorithm until its convergence. Another novelty is the use of K-means to initially assign data points to clusters. The algorithm is evaluated by comparing its performance with the conventional EM algorithm, the K-means algorithm, and the likelihood cross-validation technique on several datasets.     

基于改进粒子群优化的SVM故障诊断方法

针对支持向量机(SVM)分类模型参数选取困难的问题,提出基于遗传免疫的改进粒子群优化算法,克服传统粒子群算法前期收敛快、后期易陷入局部最优的缺陷。将该算法与优化支持向量机分类模型相结合,建立基于遗传免疫粒子群和支持向量机的诊断模型,并用于轴承故障诊断中。结果表明,基于遗传免疫粒子群算法优化的SVM可实现对SVM分类模型参数的自动优化,并能提高SVM分类模型的故障诊断精度,对分散程度较大、聚类性较差的故障样本分类有较强的适用性。     

Inverse problems

Positron emission tomography involves constructing an image of brain tissue from gamma rays counted at detectors surrounding the head. This is an inverse problem: how to measure a phenomenon from data taken from a derived distribution. We have noise and a loss of high frequency signal, both of which contribute to ill-conditioning. A brief and informal review of the mathematical and statistical methods available for handling inverse problems which are linear or linearizable is given.     

Comparison between genetic parameters of cheese yield and nutrient recovery or whey loss traits measured from individual model cheese-making methods or predicted from unprocessed bovine milk samples using Fourier-transform infrared spectroscopy

Cheese yield is an important technological trait in the dairy industry. The aim of this study was to infer the genetic parameters of some cheese yield-related traits predicted using Fourier-transform infrared (FTIR) spectral analysis and compare the results with those obtained using an individual model cheese-producing procedure. A total of 1,264 model cheeses were produced using 1,500-mL milk samples collected from individual Brown Swiss cows, and individual measurements were taken for 10 traits: 3 cheese yield traits (fresh curd, curd total solids, and curd water as a percent of the weight of the processed milk), 4 milk nutrient recovery traits (fat, protein, total solids, and energy of the curd as a percent of the same nutrient in the processed milk), and 3 daily cheese production traits per cow (fresh curd, total solids, and water weight of the curd). Each unprocessed milk sample was analyzed using a MilkoScan FT6000 (Foss, Hillerød, Denmark) over the spectral range, from 5,000 to 900 wavenumber × cm⁻¹. The FTIR spectrum-based prediction models for the previously mentioned traits were developed using modified partial least-square regression. Cross-validation of the whole data set yielded coefficients of determination between the predicted and measured values in cross-validation of 0.65 to 0.95 for all traits, except for the recovery of fat (0.41). A 3-fold external validation was also used, in which the available data were partitioned into 2 subsets: a training set (one-third of the herds) and a testing set (two-thirds). The training set was used to develop calibration equations, whereas the testing subsets were used for external validation of the calibration equations and to estimate the heritabilities and genetic correlations of the measured and FTIR-predicted phenotypes. The coefficients of determination between the predicted and measured values in cross-validation results obtained from the training sets were very similar to those obtained from the whole data set, but the coefficient of determination of validation values for the external validation sets were much lower for all traits (0.30 to 0.73), and particularly for fat recovery (0.05 to 0.18), for the training sets compared with the full data set. For each testing subset, the (co)variance components for the measured and FTIR-predicted phenotypes were estimated using bivariate Bayesian analyses and linear models. The intraherd heritabilities for the predicted traits obtained from our internal cross-validation using the whole data set ranged from 0.085 for daily yield of curd solids to 0.576 for protein recovery, and were similar to those obtained from the measured traits (0.079 to 0.586, respectively). The heritabilities estimated from the testing data set used for external validation were more variable but similar (on average) to the corresponding values obtained from the whole data set. Moreover, the genetic correlations between the predicted and measured traits were high in general (0.791 to 0.996), and they were always higher than the corresponding phenotypic correlations (0.383 to 0.995), especially for the external validation subset. In conclusion, we herein report that application of the cross-validation technique to the whole data set tended to overestimate the predictive ability of FTIR spectra, give more precise phenotypic predictions than the calibrations obtained using smaller data sets, and yield genetic correlations similar to those obtained from the measured traits. Collectively, our findings indicate that FTIR predictions have the potential to be used as indicator traits for the rapid and inexpensive selection of dairy populations for improvement of cheese yield, milk nutrient recovery in curd, and daily cheese production per cow.     

Rheological parameters determination using Near Infrared technology in whole wheat grain

Rheological properties of dough are important for wheat quality characterisation. This research sought to obtain prediction models for rheological characteristics and to characterise the breadmaking quality of whole wheat using NIRS (Near Infrared Reflectance Spectroscopy) technology, in order to offer a rapid tool to the farmers to know the quality of their product at the harvest moment. Tenacity (P), extensibility (L), deformation energy (W) and ratio P/L of dough were measured using traditional methods. NIR spectra were acquired from these samples. Models to predict the values from these parameters were developed. The SEC achieved for the extensibility, deformation energy and tenacity of dough and the ratio between the two latter parameters were 5.27 mm, 9.97 × 10⁻⁴ J, 3.98 mm and 0.025, respectively. The four models were validated by cross-validation, and by independent validation. The precision obtained in these models was enough for being applied in harvesters or at delivering moment.