Identification and Sensitivity Analysis of Complex Interaction Networks

复杂相互作用网络的识别和灵敏度分析复杂网络是由相互作用的元素与处理单元连接在一起组成的，在现代科学和工程的许多学科中无处不在。示例包括生化反应网络、蜂窝网络、流行病学网络、社会网络、组织网络、配电网络以及互联网和移动蜂窝网络。然而，大多数网络的组织结构和功能特性的知识是非常有限的。由于相互作用网络的高度复杂性，阐明相互作用网络的物理原理和结构机制是一项非常困难的任务，需要收集和系统分析大量数据。因此，人们普遍认为，迫切需要发展一种科学严谨的方法来研究相互作用网络。本研究的主要目标是开发一种通用的统计信号处理方法，用于基于模型的识别和分析来自不完整和噪声观测的复杂非线性相互作用网络。研究人员研究了严格的理论和计算技术，通过最先进的识别和模型选择方法来估计相互作用网络的结构和动态特性，并通过概率敏感性分析来研究网络的鲁棒性。为了提高计算效率，采用了两相？采用以敏感性分析为指导的网络识别方法。这种方法有效地利用了复杂交互网络对大多数参数具有鲁棒性这一事实。第一阶段的目标是从可用的测量中快速估计未知网络参数的值，而不太关心它们的准确性。第二阶段的目标是使用敏感性分析来准确估计少数“有影响”的值。通过使用选择性扰动获得的网络行为的更有信息的观察来参数。

Identification and Sensitivity Analysis of Complex Interaction NetworksComplex networks, consisting of interacting elements linked together with processing units, are ubiquitous across many disciplines of modern science and engineering. Examples include biochemical reaction networks, cellular networks, epidemiological networks, social networks, organizational networks, power distribution networks, as well as internet and mobile cellular networks. However, knowledge of the organizational structure and functional properties of most networks is very limited. Due to high complexity, elucidating the physical principles and structural mechanisms underlying interaction networks is a very difficult task, which requires collection and systematic analysis of large amounts of data. As a consequence, there is a general consensus that the development of a scientifically rigorous approach to studying interaction networks is urgently needed.The main goal of this research is to develop a general statistical signal processing methodology for model-based identification and analysis of complex nonlinear interaction networks from incomplete and noisy observations. The investigators study rigorous theoretical and computational techniques for estimating the structural and dynamic properties of interaction networks by state-of-the-art identification and model selection methodologies, and for studying network robustness via probabilistic sensitivity analysis. To achieve computational efficiency, a ?two-phase? approach to network identification is employed, guided by sensitivity analysis. This approach effectively exploits the fact that complex interaction networks are robust to most parameters. The objective of the first phase is to quickly estimate the values of the unknown network parameters from available measurements without much concern for their accuracy. The objective of the second phase is to use sensitivity analysis to accurately estimate the values of a small number of ?influential? parameters by using more informative observations of network behavior obtained by selective perturbations.