A Utility-Aware, Agent-Specific Theory of Information Latency

Abstract

The objective of the project is to design network algorithms to sense, process and transmit data over a resource-constrained wireless network, so that useful information is delivered in a timely manner to the various agents connected by the network. The state-of-the-art in information theory and network optimization theory is not adequate to design algorithms to achieve this objective because of two reasons: (i) they do not take into account spatio-temporal correlations in the collected data to minimize resource usage, and (ii) they focus primarily on steady-state performance, which does not address the dynamic nature of a mobile, ad hoc battlefield network. In this project, we take a multi-disciplinary approach involving machine learning, queueing theory, probability, edge computing, estimation, and multi-agent control theory to design network architectures and algorithms to maximize a new metric of performance called the Utility-of-Information (UoI). The UoI can be shown to generalize other metrics used in prior works to capture the usefulness of information, but more importantly, it takes into account spatiotemporal correlations, a feature not present in other metrics. To make the joint information acquisition, processing, and transport problem computationally tractable, we will appeal to a layering principle: (a) aUoI-aware end-to-end sensing layer will be designed to optimally harness data from multiple sources to construct different views of the same information that are useful to different agents in the network; (b) a theory of network architectures and algorithms will be developed to provide finite-time throughput and latency guarantees for information transport protocols; and (c) edge computing algorithms to process local information will be designed to operate in resource-constrained wireless environments. The expected outcomes are a new theory of UoI, a new theory of network architectures and protocols designed to take into the computational needs in a modern battlefield network, new scheduling algorithms for scheduling edge computing tasks, and new distributed/local algorithms for information processing at the edges of the network. The theoretical developments will be complemented with validation on a testbed. The testbed will consist of two components: a mobile ad hoc network at one university and a powerful cloud computing facility at a remote location. This will allow us to emulate DoD scenarios where a mission-specific network is supported by a remoted command center.

Document Details

Document Type
DoD Grant Award
Publication Date
Aug 20, 2019
Source ID
N000141912566

Entities

People

  • Rayadurgam Srikant

Organizations

  • Office of Naval Research
  • United States Navy
  • University of Illinois Urbana–Champaign

Tags

Fields of Study

  • Computer science

Readers

  • Adaptive Control and Estimation with Uncertainty in Dynamic Systems.
  • Computer Networking
  • Neural Network Machine Learning.

Technology Areas

  • AI & ML
  • AI & ML - Machine Learning Algorithms