Trainable fusion rules. II. Small sample-size effects

Profound theoretical analysis is performed of small-sample properties of trainable fusion rules to determine in which situations neural network ensembles can improve or degrade classification results. We consider small sample effects, specific only to multiple classifiers system design in the two-category case of two important fusion rules: (1) linear weighted average (weighted voting), realized either by the standard Fisher classifier or by the single-layer perceptron, and (2) the non-linear Behavior-Knowledge-Space method. The small sample effects include: (i) training bias, i.e. learning sample size influence on generalization error of the base experts or of the fusion rule, (ii) optimistic biased outputs of the experts (self-boasting effect) and (iii) sample size impact on determining optimal complexity of the fusion rule. Correction terms developed to reduce the self-boasting effect are studied. It is shown that small learning sets increase classification error of the expert classifiers and damage correlation structure between their outputs. If the sizes of learning sets used to develop the expert classifiers are too small, non-trainable fusion rules can outperform more sophisticated trainable ones. A practical technique to fight sample size problems is a noise injection technique. The noise injection reduces the fusion rule’s complexity and diminishes the expert’s boasting bias.     

The Effects of SNP Genotyping Errors on the Power of The Cochran-Armitage Linear Trend Test for Case/Control Association Studies

The questions addressed in this paper are: What single nucleotide polymorphism (SNP) genotyping errors are most costly, in terms of minimum sample size necessary (MSSN) to maintain constant asymptotic power and significance level, when performing case‐control studies of genetic association applying the Cochran‐Armitage trend test? And which trend test or χ² test is more powerful under standard genetic models with genotyping errors? Our strategy is to expand the non‐centrality parameter of the asymptotic distribution of the trend test to approximate the MSSN using a Taylor series linear in the genotyping error rates. We apply our strategy to example scenarios that assume recessive, dominant, additive, or over‐dominant disease models. The most costly errors are recording the more common homozygote as the less common homozygote, and the more common homozygote as the heterozygote, with MSSN that become indefinitely large as the minor SNP allele frequency approaches zero. Misclassifying the heterozygote as the less common homozygote is costly when using the recessive trend test on data from a recessive model. The χ² test has power close to, but less than, the optimal trend test and is never dominated over all genetic models studied by any specific trend test.     

Causes, Problems and Benefits of Different Between and Within Effects in the Analysis of Clustered Data

In analyzing longitudinal (panel), multicenter or otherwise clustered data, one often discovers that between and within individual or cluster regression coefficients are not equal. Clearly, this creates difficulties for interpretation and casts doubt on any causal interpretation of the coefficients. To help interpretation, it is useful to note that several types of model misspecification leading to this phenomenon can be reformulated as problems of omitted covariates. We consider models where the relationship between the outcome and the covariates is known, as long as all covariates are correctly measured and included. However, when there are unknown confounders, selection bias or measurement error, this is structurally equivalent to covariates being omitted. We first derive the correct form of the mean structure conditionally on the observed covariates, and then examine the consequences of fitting models which do not take into account the possibility of omitted covariates. We give examples of constraints, which allow estimation of the original parameters of interest. The models are used to help interpret analyses of data from the Asset and Health Dynamics of the Oldest Old (AHEAD), a panel survey of individuals age 70 and above.     

Sinusoidal forearm tracking with delayed visual feedback I. Dependence of the tracking error on the relative delay

During forearm tracking of a sinusoidally moving target matching of proprioceptive and visual feedback was altered by introducing different visual delays. In five normal subjects seven target frequencies were tested ranging from 0.3 to 1.5 Hz. For each target frequency nine different delays were used ranging from 0 to 120% of the cycle duration with target frequency and delay being varied randomly. Tracking error revealed a cyclic behaviour with an increase up to delays of about 50% of the target cycle duration and an improvement for delays larger than 50%. Modulation of response frequency was less pronounced compared with tracking error variation but also was dependent on relative phase. The response frequency matched the target frequency at delays of 0 and 100% of cycle duration and was slightly lower than the target frequency with relative delays of about 50%. The introduction of a visual delay during sinusoidal forearm tracking leads to a spatial in addition to the temporal mismatch between proprioception and vision. The temporal and the spatial incompatibility influence the tracking performance differentially. Received: 20 February 1996 / Accepted: 26 March 1997     

丝裂霉素产生菌丛生链霉菌的靶向育种

以丝裂霉素产生菌丛生链霉菌638及其衍生菌株作出发菌株,用UV诱变再经过易出误差修复和建立在丝裂霉素生物合成途径基础上的靶问育种新技术,得到许多突变型菌株。精氨酸抗性突变型菌株14-39孢子形成延迟,产生的丝裂霉素效价比原始菌株638提高7倍,丝裂霉素提取得量最高可达82mg/L。     

On the reliability of a dental OSCE, using SEM: effect of different days

Aim: The first aim was to study the reliability of a dental objective structured clinical examination (OSCE) administered over multiple days, and the second was to assess the number of test stations required for a sufficiently reliable decision in three score interpretation perspectives of a dental OSCE administered over multiple days. Materials and methods: In four OSCE administrations, 463 students of the year 2005 and 2006 took the summative OSCE after a dental course in comprehensive dentistry. The OSCE had 16–18 5‐min stations (scores 1–10), and was administered per OSCE on four different days of 1 week. ANOVA was used to test for examinee performance variation across days. Generalizability theory was used for reliability analyses. Reliability was studied from three interpretation perspectives: for relative (norm) decisions, for absolute (domain) and pass–fail (mastery) decisions. As an indicator of reproducibility of test scores in this dental OSCE, the standard error of measurement (SEM) was used. The benchmark of SEM was set at <0.51. This is corresponding to a 95% confidence interval (CI) of <1 on the original scoring scale that ranged from 1 to 10. Results: The mean weighted total OSCE score was 7.14 on a 10‐point scale. With the pass–fail score set at 6.2 for the four OSCE, 90% of the 463 students passed. There was no significant increase in scores over the different days the OSCE was administered. ‘Wished’ variance owing to students was 6.3%. Variance owing to interaction between student and stations and residual error was 66.3%, more than two times larger than variance owing to stations’ difficulty (27.4%). The SEM norm was 0.42 with a CI of ±0.83 and the SEM domain was 0.50, with a CI of ±0.98. In order to make reliable relative decisions (SEM <0.51), the use of minimal 12 stations is necessary, and for reliable absolute and pass–fail decisions, the use of minimal 17 stations is necessary in this dental OSCE. Conclusions: It appeared reliable, when testing large numbers of students, to administer the OSCE on different days. In order to make reliable decisions for this dental OSCE, minimum 17 stations are needed. Clearly, wide sampling of stations is at the heart of obtaining reliable scores in OSCE, also in dental education.     

不同器材在制备菌片中所滴10μl菌液菌量误差的试验观察

经试验，用注射器、２０μｌ移液器与０．１ｍｌ微量吸管所滴枯草杆菌黑色变种芽胞菌悬液（１０μｌ），回收菌量在平均值±１０％范围内者分别占样本数的８６．６％、８５．０％与４１．７％。其可重复性，以前两者为优。     

可调恒压式毛细管粘度计误差分析

本文介绍了可调恒压式毛细管粘度计的工作原理，详尽分析了弯液面附加压强和驱动压差不稳定引起的误差及减少误差的方法。对有关检测人员和医师正确使用这种粘度计和科学地分析，使用测量结果有指导意义。     

Sporadic late onset ornithine transcarbamylase deficiency in a boy with somatic mosaicism for an intragenic deletion

Somatic mosaicism for a deletion in the ornithine transcarbamylase gene is described in a boy with sporadic late onset ornithine transcarbamylase deficiency. These findings are discussed in relation to the clinical picture of the patient and in relation to genetic counseling.     

Cerebellar aminergic neuromodulation: towards a functional understanding

Although a number of neuromodulators influence the cerebellar circuitry, their functions remain largely unknown. By reviewing and combining results from data-driven and theory-driven studies, we attempt to provide an integrated systems view of cerebellar neuromodulation. First, we review the short- and long-term effects of neuromodulators on the cerebellar circuitry. Second, we review recent theories of the cerebellum and show that a number of modulatory signals are needed for powerful cerebellar learning and control. Finally, we attempt to match each theoretically derived modulatory signal with a specific neuromodulator. In particular, we propose that serotonin controls the ‘responsibility’ of each cerebellar unit (or microcomplex) in cerebellar learning and control; norepinephrine gates unsupervised learning in the cerebellar cortex; dopamine enhances goal-oriented cerebellar learning; and, finally, acetylcholine controls the speed of supervised learning in Purkinje cells.