CIF: Small: Data Reduction for Networked Inference

设备、计算、通信和联网方面的技术进步已经允许开发和部署具有不同应用的不同规模的联网感测系统。然而，在这些网络系统中收集的大量数据提出了一些独特的挑战。由于这种系统固有的各种限制，在大规模网络中很难确保有用的数据及时地在正确的位置可用。此外，收集到的原始数据往往不够清晰，无法做出知情决定。该项目的主要目标是开发理论和方法，使原始数据减少到这样的程度：1）它们传输到所需的目的地是符合系统的限制，2）没有丢失的信息有关的使命的网络系统。该项目追求在这样的推理网络中进行数据约简的综合处理，旨在发展充分性原则和相关的理论和方法来克服这些挑战。在网络环境中，推理数据简化和传输数据压缩之间的紧密联系得到了利用和扩展。一些经典的多终端信源编码问题的充分性原理的新的解释将被开发，他们提供了一个新的场所，探索统计和信息理论之间的联系。在网络推理系统中指导数据简化的理论和方法的发展也有可能统一针对特定推理问题定制的现有工作。

Technological advances in devices, computing, communication, and networking have allowed the development and deployment of networked sensing systems of varying scales with diverse applications. The voluminous data collected in these networked systems, however, present some unique challenges. Ensuring that useful data be available at the right location in a timely fashion is difficult in a large scale network because of various limitations inherent in such systems. In addition, the collected data in its raw form often lacks clarity for making an informed decision. The primary goal of the project is to develop theory and methodology that allow the raw data to be reduced to the extent such that 1) their transmission to the desired destination is compliant to the system limitation, and 2) there is no loss of information with respect to the mission of the networked system. The project pursues a comprehensive treatment of data reduction in such inference networks and aims to develop the sufficiency principle and related theory and methods to overcome these challenges. Intimate connection between data reduction for inference and data compression for transmission is exploited and expanded in the network setting. New interpretations using the sufficiency principle for some classical multi-terminal source coding problems will be developed; they provide a new venue for the exploration of the connection between statistics and information theory. The development of theory and methods that guide data reduction in a networked inference system also has the potential to unify existing works that are tailored toward specific inference problems.