Research

Federated Resource Management

Federated learning is a novel approach to distributed machine learning under the MEC framework. Each edge edge cloud is given a copy of some machine learning model \(f(\cdot)\) that is trained globally by the central cloud. The individual edge clouds then update the parameters of their copy of the model by training it on their data that they interact with. Then, over time, the central cloud aggregates the training of each individual edge cloud to update the global model.

This project proposes to use concepts of federated learning and reinforcement learning to provide optimal solutions for service placement and request scheduling in multi-tier edge and cloud computing frameworks. This project will focus on new algorithms and system design technologies to trade off the quality-of-service (QoS) and the cost in edge computing systems. Three main thrusts include: 1) finding theoretical bounds through mathematical formulation and algorithm design, 2) finding more practical and adaptive solutions through multi-agent reinforcement learning and the federated learning framework, and 3) comparison of practical solutions with theoretical bounds through numerical analysis and implementations on a mobile edge computing test-bed.

---

Mobile Edge Computing

Mobile edge computing (MEC) is a cloud computing framework where a large number of cheaper, less powerful devices, known as edge clouds, are dispersed in a given area to provide more immediate attention to requests made by users' mobile devices.

Given a multi-tier M:N edge-to-cloud computing framework, predefined resource constraints on the edge clouds, predefined set of QoS requirements for  different  users, user mobility,and a set of policies on different clouds, we are  interested  in finding optimal coded based and non-coded service  placement, and service migration strategies as well as task scheduling strategies that trade-off the total cost of the system with the provided QoS for the users in the system.

Relevant Papers:

• It's hard to share: joint Service placement and request scheduling
• QoS-Aware Placement of Deep Learning Services on the Edge with Multiple Service Implementations

---

Social Network Analysis

Social networks can be found everywhere in our everyday lives. Online social networking platforms (e.g., Facebook, Twitter) provide platforms for information and disinformation to spread and affect social happenings. In this line of research, we approach the phenomenon of how information, influence, and opinion diffuse in social networks via social communication. In social science literature, the ways in which people engage with information is often influenced by their own behavioral predispositions and personally-held beliefs. As such, there is a need to consider such factors when modeling how information (or disinformation) can spread in socail network systems. Our research in this area aims to make such considerations when studying how information, influence, and opinion can spread on these platforms.

---