This paper summarizes the research on population-based probabilistic search algorithms based on modeling promising solutions by estimating their probability distribution and using the constructed model to guide the further exploration of the search space. It settles the algorithms in the field of genetic and evolutionary computation where they have been originated. All methods are classified into a few classes according to the complexity of the class of models they use. Algorithms from each of these classes are briefly described and their strengths and weaknesses are discussed.

Abstract In this paper we introduce model-based search as a unifying framework accommodating some recently proposed heuristics for combinatorial optimization such as ant colony optimization, stochastic gradient ascent, cross-entropy and estimation of distribution methods. We discuss similarities as well as distinctive features of each method, propose some extensions and present a comparative experimental study of these algorithms. 1

Meta-analysis collects and synthesizes results from individual studies to estimate an overall effect size. If published studies are chosen, say through a literature review, an inherent selection bias may arise, since for example, studies may tend to be published more readily if they are statistically significant, or deemed to be more `interesting' in terms of the impact of their outcomes. We develop a simple rank-based data augmentation technique, formalizing the use of funnel plots, to estimate and adjust for the numbers and outcomes of missing studies. Several non-parametric estimators are proposed for the number of missing studies, and their properties are developed analytically and through simulations. We apply the method to simulated and epidemiological data sets, and show it is both effective and consistent with other criteria in the literature. Corresponding author's email address: tweedie@stat.colostate.edu Key words: Meta-analysis; Publication bias; Missing studies; File dra...

Experimental comparisons between statistical and machine learning methods appear with increasing frequency in the literature. However, there does not seem to be a consensus on how such a comparison is performed in a methodologically sound way. Especially the effect of testing multiple hypotheses on the probability of producing a "false alarm" is often ignored. We transfer multiple comparison procedures from the statistical literature to the type of study discussed in this paper. These testing procedures take the number of tests performed into account, thereby controlling the probability of generating "false alarms". The multiple comparison procedures selected are illustrated on well-known regression and classification data sets. 26.1 Introduction Recent interactions between the statistical and artificial intelligence communities (see e.g. [Han93, CO94]), have led to many studies that compare the performance of empirical statistical and machine learning methods on real-life data sets; ...

Abstract not found

this article unless noted otherwise). The names of ss'mptoms are abbreviated because of space limitations (full names of the criterial symptoms can be fbund in Table 1). The symptoms circled with double lines are criterial symptoms, and those circled with single lines are characteristic symptoms. One interesting result to notice is that even among criterial symptoms, causal centrality (as grossly indicated by the number of symptoms that a symptom causes and their strengths) seems highly variable. For instance, on average, participants believe that in anorexia nervosa, "fear of being fat even when underweight" causes many symptoms, including fear of eating in public, bingeing and purging, excessive dieting, and refusal to gain weight. However, "absence of the period (in women) for 3+ menstrual cycles," another criterial symptom for anorexia nervosa, was rarely judged to cause any other symptoms of that disorder. Figure 1 suggests that a conceptually central symptom (e.g., "fear of being fat even when underweight" in anorexia nervosa) is also causally central, and a conceptually peripheral symptom (e.g., "absence of the period (in women) tbr 3+ menstrual cycles") is also cansally peripheral. This would be consistent with our primary hypothesis that a symptom is conceptually central to the extent that it causes other features. However, it is also apparent from Figure 1 that not all such pairwise comparisons, as in all of psychological research, were perfect one-to-one correspondences. Therefore, we tested for statistical significance as follows

Neuronal activity was recorded extracellularly from 20 pop-ulations in the lateral cortex of the left anterior temporal lobe of 11 patients undergoing awake craniotomy for epilepsy, during an input-distraction-retrieval measure of recent ver-bal memory that also included two later successive retrievals of the same information after additional distracting tasks. Changes in activity were determined for each 1 set epoch in three major comparisons: (1) the same visual cues used for naming an input to recent memory, naming without a memory component, and a spatial matching task; (2) memory input (Ml), distraction (S), and initial cued retrieval (Rl) from memory, where object naming was the input to memory and naming of other objects the distracters; (3) initial retrieval (Rl) and the two subsequent serial retrievals of the same information (R2, R3). Control comparisons were also made

Following the framework in Oliva, DeSarbo, Day, and Jedidi (1987), GEMCAT II implements a flexible method to test catastrophe models containing multivariate (i.e., latent) variables while allowing for a priori variable specifications. The system uses an efficient hybrid minimization algorithm combining the Downhill Simplex and Powell's Conjugate Gradient method. GEMCAT II is compiled in Delphi V3.0 and is sufficiently fast to allow for the use of resampling methods (bootstrap as well as jackknife) to determine the statistical significance of latent variables' indicator weights. In addition, a Pseudo-R 2 index of model fit is provided, together with a test of significance, and options are included facilitate competitive model tests of nested and non-nested catastrophe models as well as linear models. Two simulation studies are reported. Based on 61,250 simulated data sets of varying sizes, the first study addressed the effects of indicator reliability on the quality of the weight estimations, while the second dealt with the problem of "false positives" in model identification. The results strongly support the viability of the GEMCAT II approach over a wide range of reasonable indicator reliabilities and sample sizes. Moreover, it proved possible to distinguish reliably between cusp catastrophes and linear models based on the Pseudo-R 2 values. Finally, GEMCAT II is applied to actual market data in order to demonstrate its use in an economic context. Using 34 quarters of panel data, we examine the fit of a cusp catastrophe model of organizational product adoption as applied to competing software standards in the presence of network externalities. The results are consistent with economic theory and published work on network externalities. This example also illustrate...

Abstract not found

This report is the result of a strong cooperation between all partners of the project. The different topics of the benchmarking methodology have been the subject of several useful discussions during the partners meetings held in Lausanne, Barcelona and Louvain-La-Neuve (on the occasion of the Elena Industrial Workshop) during the last year of the project. Moreover the coordination of the benchmark task was subject to a constructive bilateral cooperation between the INPG and UCL partners (two one-week visits). The main objectives of task B4 may be summarized as follows: