2008 Workshop on Bayesian Model Selection and Objective Methods

模型选择在频率设置是一个很发达的领域。在贝叶斯框架中，原则上，模型选择是非常简单的。先验概率分布用于描述围绕所有未知数的不确定性，包括正在考虑的模型和这些模型的参数。观察数据后，后验分布为剩余的不确定性提供了一个连贯的数据后总结，这与模型选择有关。然而，这种方法的实际实现并不简单，并且涉及诸如先验选择、可解释性和计算可行性等问题。在本次研讨会上，12位在贝叶斯模型选择领域工作的杰出人士介绍了他们在许多不同领域的工作，包括确定良好的客观先验，评估各种信息标准（AIC， BIC， RIC），贝叶斯因子的计算方法以及马尔可夫链蒙特卡罗。一些年轻的研究人员参加了研讨会，并在海报会议上展示了他们的工作。变量选择是科学和医学研究中一个重要而普遍的问题。一些重要的变量将从许多候选变量中选择出来，用于理解、预测和决策。从历史上看，变量选择是在频率设置中进行的。然而，贝叶斯方法提供了重要的优势。从广义上讲，它们提供了一种连贯的方法来处理预测变量已知的个体未来反应的分布。计算能力和统计方法的最新进展大大提高了贝叶斯方法在回归和变量选择方面的可行性。研讨会提供了一个很好的机会来讨论贝叶斯模型选择和客观方法的许多最近的重大发展；讨论已经发现哪些是有效的，哪些是无效的；并找出重要的问题和新的研究方向。

Model selection in the frequentist setting is a well developed field. In the Bayesian framework, in principle, model selection is very simple. Prior probability distributions are used to describe the uncertainty surrounding all unknowns, including models being considered and the parameters for these models. After observing the data, the posterior distribution provides a coherent post data summary of the remaining uncertainty which is relevant for model selection. However, the practical implementation of this approach is not straightforward, and involves issues such as choice of priors, interpretability, and computational feasibility. In this workshop, twelve distinguished individuals who work in Bayesian model selection present their work in a number of different areas, including determination of good objective priors, assessment of various information criteria (AIC, BIC, RIC), methods of calculation of Bayes factors, and Markov chain Monte Carlo. A number of young researchers participate in the workshop and present their work in poster sessions.Variable selection is an important and pervasive problem in scientific and medical research. A few important variables are to be selected from many candidates and used for understanding, prediction and decision making. Historically, variable selection has been carried out in a frequentist setting. However, Bayesian approaches offer important advantages. In broad terms, they give a coherent way of dealing with the distribution of the future response of an individual for whom the predictor variables are now known. Recent advances in both computing power and statistical methodology have greatly enhanced the feasibility of Bayesian approaches to regression and variable selection. The workshop provides an excellent opportunity to discuss the many recent significant developments in Bayesian model selection and objective methods; to discuss what has been found to work and what does not; and to identify important problems and new research directions.