Collaborative Research: NeTS: Small: CoLeNe: Cooperative Learning in Heterogeneous Edge Networks

Cooperation of multiple devices to learn and make decisions based on their environment is especially valuable for Internet of Things (IoT) and other applications. Existing algorithms for cooperative learning often assume that all devices face the same set of decision choices, which is not often the case in networked system settings. This project, CoLeNe (Cooperative Learning in heterogeneous Networks), proposes to design and evaluate algorithms for multiple devices to cooperatively learn for decision making over a large set of choices in computer networks as the agents may face different sets of decisions. The developed algorithms allow devices to collaboratively explore the decision space and identify the best option from a large set faster. The project also applies these algorithms to the decision-making cases in networks and demonstrates their usefulness through the examples. In addition, the research effort is paired with educational and outreach initiatives that introduce students to the theory and practice of cooperative learning in networks. This proposal aims to develop cooperative online learning algorithms that are communication-efficient and robust to heterogeneity in computation, data, and privacy across agents. It consists of two research thrusts: (1) theoretical foundations and algorithms. The project will extend existing theoretical work on online learning to the settings of multiple heterogeneous agents with different sets of decision choices and privacy constraints on the information they can share with other agents. It develops algorithms for multiple devices to cooperatively learn to make optimal decisions. (2) Implementing two network applications. The developed cooperative online learning algorithms will be adapted to two applications in edge networks: distributed placement of computing applications across a network of devices, and optimization of wireless network configurations and transmission schedules. The learning framework developed through this project is expected to have broad applicability across IoT as well as other domains involving distributed, heterogeneous learning systems. 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.