天文数据集中离散目标检测的快速贝叶斯方法-- PowellSnakes I
A fast Bayesian approach to discrete object detection in astronomical datasets - PowellSnakes I
摘要:
提出了一种新的快速贝叶斯方法,用于漫反射背景中离散目标的检测。这种称为PowellSnakes的新方法通过以下方式加速了传统的贝叶斯方法:i)用一种更快的精确形式代替表征离散对象的参数的标准似然形式:ii)使用基于Powell方向集算法的同时多重最小化代码来快速定位后验中的局部极大值;以及iii)通过使用基于峰值的局部高斯近似的近似证据值执行贝叶斯模型选择来判定每个定位的后验峰值是否对应于真实对象。该高斯近似的构造还提供了所讨论对象的导出参数值中的不确定性的协方差矩阵。这种新的方法提供了一个加速性能的一个因素的“数百”相比,现有的贝叶斯源提取方法,使用MCMC探索参数空间,如霍布森和麦克拉克伦。我们说明了该方法的能力,适用于一些简化的玩具模型。此外,PowellSnakes具有基于先验一致地定义接受/拒绝的阈值的优点,这在频率论方法中是不能说的。我们在这里提出的第一个实现这种技术(版本-I)。目前正在研究如何进一步改进这一执行工作,不久将公布结果。将在即将出版的出版物中介绍该方法在现实模拟普朗克观测中的应用。
英文摘要:
A new fast Bayesian approach is introduced for the detection of discrete objects immersed in a diffuse background. This new method, called PowellSnakes, speeds up traditional Bayesian techniques by: i) replacing the standard form of the likelihood for the parameters characterizing the discrete objects by an alternative exact form that is much quicker to evaluate; ii) using a simultaneous multiple minimization code based on Powell's direction set algorithm to locate rapidly the local maxima in the posterior; and iii) deciding whether each located posterior peak corresponds to a real object by performing a Bayesian model selection using an approximate evidence value based on a local Gaussian approximation to the peak. The construction of this Gaussian approximation also provides the covariance matrix of the uncertainties in the derived parameter values for the object in question. This new approach provides a speed up in performance by a factor of `hundreds' as compared to existing Bayesian source extraction methods that use MCMC to explore the parameter space, such as that presented by Hobson & McLachlan. We illustrate the capabilities of the method by applying to some simplified toy models. Furthermore PowellSnakes has the advantage of consistently defining the threshold for acceptance/rejection based on priors which cannot be said of the frequentist methods. We present here the first implementation of this technique (Version-I). Further improvements to this implementation are currently under investigation and will be published shortly. The application of the method to realistic simulated Planck observations will be presented in a forthcoming publication.
机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。