Compressive Sensing Based Inference in Distributed Networks

本项目的目标是在资源受限的分布式网络中，通过基于压缩感知的测量方案获得降维子空间中的分散推理，并为其开发有效的算法和获得性能边界。该方案的优点在于，它将压缩感知理论推进到具有多个测量向量的资源受限分布式系统中。虽然文献中对基于压缩感知的单测量向量信号处理进行了大量的研究，但在资源受限的分布式系统中，将其扩展到多测量向量仍处于起步阶段。所提出的研究的变革性质在于它具有识别和解决基于低维采样方案的分散推理问题的潜力，该方案利用在分布式节点收集的多个测量向量的结构化模型。更广泛的影响拟议的研究将对推动理论和实践，以及教育和培训产生直接影响。更广泛的影响是：1)开发高效的低维信号处理技术，用于在资源受限的分布式网络中进行分散推理，利用大型数据集的结构化属性（稀疏性以及超越简单稀疏性模型的模型）。2)具有广泛应用的分布式推理网络的高效设计方法，包括环境监测、遥感、雷达网络、认知无线电网络和具有高维数据的生物医学应用。3)开发研究生水平的大数据集信号处理和推理课程。指导研究生的研究方法。

ObjectiveThe objective of this program is to develop efficient algorithms and obtain performance boundsfor decentralized inference in the reduced dimensional subspace obtained via Compressive Sens-ing based measurement schemes in resource-constrained distributed networks.Intellectual MeritThe intellectual merit of the proposed program is that it will advance the theory of CompressiveSensing to resource-constrained distributed systems with multiple measurement vectors. Whilethere is a considerable amount of work in the literature for Compressive Sensing based signalprocessing with a single measurement vector, its extension to multiple measurement vectors inresource-constrained distributed systems is still at its infancy. The transformative nature of theproposed research lies in its potential to identify and solve decentralized inference problems basedon low dimensional sampling schemes exploiting structured models of the multiple measurementvectors collected at distributed nodes.Broader ImpactsThe proposed research will have direct impact on advancing the theory and practice, as well ason education and training. The broader impacts are: 1) Development of efficient low dimensional signal processing techniques for decentralized inference in resource-constrained distributed networks exploiting structured properties (sparsity as well as models that go beyond the simple sparsity model) of large data sets. 2) Efficient design approaches for distributed inference networks with a wide variety of applications including environmental monitoring, remote sensing, radar networks, cognitiveradio networks and bio-medical applications with high dimensional data. 3) Development of graduate level courses on signal processing and inference for large data sets. Mentoring of graduate student in research methods.