CIF:Small:Submodular Optimization Techniques for Sensing, Signal Processing, and Inference.

越来越多的现代系统依靠传感器收集测量值，通过适当设计的信号处理从这些测量值进行推断。其中包括自动驾驶车辆、环境监测系统、智能建筑、安全和监视系统、空中交通管制等。此外，在许多现有的和新兴的应用中，在系统设计中通常存在必须考虑的严格的能量、成本、带宽或其他限制。传统上，传感器的布局、调度和功能问题一直是困难的优化问题。这项研究发展了子模块优化技术在解决此类问题中所起的重要作用。事实上，尽管近年来在子模优化工具和应用方面有了很大的发展，但涉及传感、信号处理和推理的大量应用仍有待开发。研究调查了这些问题的几个实例和方面，包括：受各种实际约束的被动和主动传感系统中的空间阵列的优化；传感器布置和调度的联合优化，以有效地学习环境动力学；受感兴趣的推理任务的相关信号处理的计算约束的空间和时间传感的优化；针对涉及高级和低级别推理的目标应用的子模信息量度的发展；开发健壮的、自适应的和通用的传感设计，以减轻系统和环境各方面的不完全知识的影响；以及通过在没有凸性的情况下使用子模块的加权选择公式来优化超出空间和时间的传感系统参数。这些调查中的每一项都涉及在当代激励应用程序的指导下，开发所需的分析、算法和体系结构。

A rapidly growing collection of modern systems rely on sensors to collect measurements from which inference is performed via suitably designed signal processing. These include autonomous vehicles, environmental monitoring systems, smart buildings, security and surveillance systems, air traffic control, and many others. Moreover,in many of these existing and emerging applications, there are often stringent energy, cost, bandwidth, or other constraints that must be taken into account in the system design. Problems of placement,scheduling, and functionality of sensors have traditionally been difficult optimization problems. This research develops the important role that submodular optimization techniques have to play in addressing such problems. Indeed, while there has been significant development in submodular optimization tools and applications in recent years, a wealth of applications involving sensing, signal processing and inference have yet to be explored.The research investigates several instances and aspects of these problems, including: optimization of spatial arrays in passive and active sensing systems subject to diverse practical constraints; joint optimization of sensor placement and scheduling for efficient learning of environmental dynamics; optimization of spatial and temporal sensing subject to computational constraints on the associated signal processing for the inference task of interest; development of families of submodular information measures for target applications involvingboth high- and low-level inference; development of robust, adaptive, and universal designs for sensing that mitigate the effects of incomplete knowledge of aspects of the system and environment; and optimization of sensing system parameters beyond space and time via weighted selection formulations exploiting submodularity in the absence of convexity. Each of these investigations involves the development of the required analysis, algorithms, and architectures, guided by contemporary motivating applications.