CAREER: Sparse Graph-Based Codes for Network Data Compression

点击翻译按钮获取中文摘要

Driven by emerging systems, such as the Internet-of-things (smart homes, wearables, connected cars, and so on), society is generating and using massive amounts of data at an ever increasing rate. For example, the amount of data created over the next three years is predicted to be more than that created over the past 30 years. Without significant technological advances, existing communications infrastructure will not be able to cope with this exponential increase. This project addresses this challenge by exploring how such data can be compressed over networks to significantly reduce the traffic that needs to be transmitted. The main idea is to explore new data-compression schemes that leverage untapped gains by exploiting similarities and structure in the data as well as how the devices are connected in the network. Examples include transmitting many related measurements in different locations of the power grid to a single destination, or transmitting a video replay to the individual devices of a large crowd in a sports stadium. As such, the proposed research promises to provide a significant transformative impact on many critical applications employing reliable networked data compression, for example in the fields of healthcare, environmental monitoring, and finance. The project also includes an integrated education plan to increase participation in Science, Technology, Engineering, and Mathematics (STEM), particularly among minority groups. This objective is supported by several complementary initiatives, including targeted K-12 activities as well as related teacher training and mentoring.The proposed research significantly advances the state of the art in network data compression by employing ideas from network coding, graph theory, iterative information processing, machine learning, and circuit design. The project involves several fundamental themes related to network-aware, low-complexity, and throughput-efficient data compression schemes which are not present in previous studies: a theoretical analysis and design of general schemes for lossy source coding, involving a characterization of finite-length scaling properties under message passing encoding and analysis of harmful graphical substructures; an investigation of the fundamental rate-distortion performance of nested graph-based codes in canonical network structures, exploring the achievable network gains in compression for practical spatially coupled constructions; and algorithmic advances and novel high-speed field programmable gate array hardware architectures. The theoretical results of this project have the potential to advance our fundamental understanding of coding strategies to achieve network gains in source compression, opening new opportunities and challenges.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.