Khamfroush, Hana

NetScience Lab overview

Director: Dr. Hana Khamfroush

Research overview: The NetScience research lab is focused on modeling and analyzing different large-scale networks and systems including cyber-physical-social systems, edge and cloud computing systems, and social networks. We are working on building the next generation of our critical infrastructures more robust, more reliable, more secure and more efficient through real-time smart decisions. Humans are becoming a part of the smart cities, where the Internet-of-things (IoT) devices such as connected sensors, traffic lights, and meters are used to collect and analyze data to control and improve infrastructures, public utilities and services to the customers. Therefore, understanding the interactions between humans, smart devices, and the infrastructures will be helpful to design more meaningful control strategies. Our goal is to understand such interactions and use this understanding to make better design and control decisions. More specifically, we focus on efficiently analyzing different data sources such as online social network and IoT (Internet of Things) devices to predict and control events in our critical infrastructures. Our mission is to build a distributed platform that can optimally and efficiently analyze the big data collected from different data sources to design more reliable, more efficient, and more robust interdependent infrastructures for smart cities.

Integrated and end-to-end machine learning pipeline for edge-enabled IoT systems: a resource-aware and QoS-aware perspective

We have recently started a project which requires access to GPU for running large language models. The work is on automated federated data analytics and for a part of this project we will need to tune and maybe retrain LLM models. LLM models are usually very heavy since they are coming with billions of parameters, therefore, we will need resources to store these models and be able to train them. We will use off-the-shelf and open access models and retrain them for our purpose of study.

Students:

Mahanipour, Afsaneh

Hosseinzadeh, Minoo 

Federated Mobile Edge Computing

Cloud computing is a general framework to provide powerful computational resources to a large number of people to meet a diverse set of demands. Mobile edge computing (MEC) is a cloud computing framework where a large number of cheaper, less powerful devices are dispersed in a given area to provide more immediate attention to requests made by users' mobile devices. These devices that handle user requests in MEC are known as edge clouds (or cloudlets). Federated learning (FL) is a novel approach to distributed machine learning under the MEC framework. Each edge cloud is given a copy of some machine learning model f(⋅) 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 is interested in studying federated mobile edge computing (F-MEC) under the consideration of a multi-tier MEC infrastructure.

Students:

Postdocs

Ansary, Aram

Hosseinzadeh, Minoo

Hudson, Nathaniel

Undergraduates:

Moser, Susanna

Computational methods:

Simulation, model training.

Softwares:

Python 3.6+, PyTorch, Tensorflow, C++, C, Socket Programming Tools, Flower.

Software Availability: All software listed are open source and freely available.

QoS-Aware Resource Management in a Three-Tier Edge Computing Framework

In this work, we study a three-tier edge-to-cloud testbed to investigate problems related to service placement and offloading schemes in edge computing systems. The testbed is implemented using a Linux computer (Intel Core i5-3470-3.20 GHz, RAM 8 GB), two Raspberry Pi 4s (Quad core, RAM 4 GB), and three Raspberry Pi 3Bs (Quad Core, RAM 1GB). The Linux computer is used as the cloud server and Raspberry Pi 4s plays the edge servers’ role. To imitate the long delay between the cloud server and the edge servers, we used one Raspberry Pi 3 as a forwarder in cooperation with one router (NetGear R6020). The two Raspberry Pis 3 are users in this system. We investigate the resource management problem in the three-layer edge system. Then, we have since added multiple Cisco routers to our testbed. Currently, we are working on resource management problems in the three-layer edge system with heterogeneous edge servers.

Students:

Postdocs

Ansary, Aram

Hosseinzadeh, Minoo

Undergraduates:

Wachal, Andrew

Computational methods:

Simulation, model training, Emulation.

Softwares:

Python 3.6+, PyTorch, Tensorflow, C++, C, Socket Programming Tools.

Software Availability: All software listed are open source and freely available.

Smart traffic control for autonomous and self-driving vehicles

The next generation of vehicles, also called, autonomous vehicles are capable of sensing the environment and moving safely with limited or no human interaction. This, however, requires a large set of smart decisions to be made in real-time and given the huge data collected by these vehicles, the challenge is how to process such data in real-time in order to make smart and reliable decisions. This work includes applications of AI and distributed learning techniques in making smart decisions and designing useful algorithms for self-driving cars.

Students:

Postdocs

Ansary, Aram

Hudson, Nathaniel

Computational methods:

Simulation, model training,

Softwares:

SUMO (Simulation of Urban Mobility), TraCI, Ray and RlLib, Python 3.6+

Software Availability: All software listed are open source and freely available.

Grants:

Khamfroush, Hana, CRII: CSR: Federated Resource Management in Mobile Edge Computing, funded by the NSF, March 2020 - Feb 2022

Khamfroush, Hana, REU supplement, CSR: REU Supplement to “CRII: CSR: Federated Resource Management in Mobile Edge Computing

Khamfroush Hana, Cisco Research Inc. Optimal service placement for mobile users in multi-tier edge-to-cloud computing framework, Jan 2020-Jul 2021

Publications: