| Project Id
|
BITSRMIT024B001276
|
| Project Detail
|
| Project Title
|
Optimizing Systems for Edge-Based Machine Learning Models
|
| Senior Supervision Team (BITS)
|
| Supervisor name and Title
|
Arnab K. Paul
|
School or Department (or company, if applicable)
|
BITS PILANI, GOA CAMPUS
|
|
|
|
Email ID
|
arnabp@goa.bits-pilani.ac.in
|
| URL for more info
|
https://arnabkrpaul.github.io/
|
| a) Are you currently supervising a BITS or RMIT HDR student?
|
YES
|
| Please comment how many you are supervising
|
2
|
| b) Have you supervised an offshore candidate before?
|
NO
|
| If no, what support structures do you have in place?
|
|
| If yes, please elaborate
|
N
|
| Senior Supervision Team (RMIT)
|
| Supervisor name and Title
|
Malka N. Halgamuge
|
School or Department (or company, if applicable)
|
COBL
|
|
|
|
Email ID
|
malka.halgamuge@rmit.edu.au
|
| URL for more info
|
https://www.rmit.edu.au/contact/staff-contacts/academic-staff/h/halgamuge-dr-malka
|
| a) Are you currently supervising a BITS or RMIT HDR student?
|
NO
|
| Please comment how many you are supervising
|
|
| b) Have you supervised an offshore candidate before?
|
NO
|
| If no, what support structures do you have in place?
|
|
| If yes, please elaborate
|
|
| Other Supervisors (BITS)
|
| Supervisor name and Title
|
|
School or Department (or company, if applicable)
|
|
| Phone Number (Optional)
|
|
Email ID
|
|
| URL for more info
|
|
| Other Supervisors (BITS)
|
| Supervisor name and Title
|
Prof Afreen Huq
|
School or Department (or company, if applicable)
|
COBL
|
| Phone Number (Optional)
|
+61399255198
|
Email ID
|
afreen.huq@rmit.edu.au
|
| URL for more info
|
https://www.rmit.edu.au/profiles/h/afreen-huq
|
| Field of Research (For Codes)
|
| 460603 | Cyberphysical systems and internet of things | 40.00 |
| 461199 | Machine learning not elsewhere classified | 30.00 |
| 490304 | Optimisation | 30.00 |
|
| Project Description
|
|
Edge-based machine learning (ML) models are essential for real-time data processing and decision-making in drones, self-driving cars, and healthcare. Unlike traditional centralized ML models, these models operate in a distributed manner, processing data across heterogeneous edge devices (smartphones or single-board computers) and networks (wireless or wired or bluetooth), posing challenges regarding data privacy, system heterogeneity, and scalability.
This project proposes a novel framework to address these challenges, enabling efficient deployment of distributed ML models on edge devices, including data privacy-aware federated learning (FL) models. The framework will incorporate edge devices, models, and datasets' heterogeneous nature while ensuring data privacy and providing a plug-and-play approach for non-technical users. The optimizations intended in the framework will also contribute to energy usage minimization so that battery-operated edge devices can be used optimally.
The key objectives are: (1) Develop a framework that integrates distributed ML and FL models, accounting for device heterogeneity and data privacy; (2) Evaluate the framework's performance on diverse use cases and experimental testbeds; (3) Derive a mathematical model to characterize the relationship between edge device parameters, ML models, and datasets.
The methodology involves: (1) Conducting a comprehensive literature review of state-of-the-art distributed ML and FL techniques; (2) Deploying a testbed to run experiments with diverse datasets, collecting time-series data on system usage and model performance; (3) Formulating a mathematical model based on the collected data; (4) Developing the proposed framework; (5) Testing the framework on real-world use cases in healthcare and drone applications.
This research will contribute to the advancement of edge computing by providing a framework that enables efficient deployment of distributed ML models on heterogeneous edge devices while ensuring data privacy and scalability, ultimately enhancing real-time decision-making capabilities in critical domains.
|
| Project Deliverable/Outcomes
|
|
(1) A comprehensive mathematical model that characterizes the relationship between heterogeneous edge device parameters, distributed machine learning (ML) models, and diverse datasets. This model will provide insights into optimizing model performance, resource utilization, and data privacy across edge devices.
(2) A comprehensive dataset containing time-series data on system usage and model performance metrics collected from experimental testbeds. This dataset will serve as a valuable resource for further research and benchmarking in edge computing and distributed ML.
(3) An open-source framework that integrates distributed ML and federated learning (FL) models, accounting for device heterogeneity and data privacy concerns. This framework will offer a plug-and-play approach, enabling non-technical users to deploy and manage distributed ML models on edge devices efficiently.
(4) Validation of the proposed framework through evaluation of a real-world application, such as an ambulance dispatch system in a hospital setting or real-time delivery decisions using drones. This evaluation will demonstrate the framework's efficacy in enhancing real-time decision-making capabilities in critical domains.
(5) Three high-quality publications in prestigious journals and/or conferences, disseminating the research findings, mathematical models, and the proposed framework to the broader scientific community. These publications will contribute to advancing edge computing and distributed ML technologies.
(6) We are in talks with industries who are interested in this work on systems for edge ML. Therefore, this project will forge industry collaborations.
|
| Research Impact Themes
|
| ADVANCED DIGITAL TECHNOLOGIES AND BUSINESS TRANSFORMATION | ARTIFICIAL INTELLIGENCE AND MODELLING; SUPPLY CHAINS AND OPTIMISATION |
| ADVANCED DIGITAL TECHNOLOGIES AND BUSINESS TRANSFORMATION | DEEP LEARNING AND PREDICTIVE MODELLING |
| SUSTAINABLE DEVELOPMENT AND ENVIRONMENT
| SUSTAINABLE TECHNOLOGIES |
|
| Which RMIT Sustainable Development Goal (SDG) does your project align to
|
|
INDUSTRY, INNOVATION, AND INFRASTRUCTURE
|
| Which RMIT Enabling Impact Platform (EIP) does your project align to
|
|
SUSTAINABLE TECHNOLOGIES AND SYSTEMS PLATFORM
|
| Which RMIT Program code will this project sit under?
|
|
DR201 (BUSINESS INFORMATION SYSTEMS)
|
| Student Capabilities and Qualifications
|
|
Coding proficiency in Python/C/C++/Java
|
|
Experience with Machine Learning/AI
|
|
M.Tech. in Computer Science (or any allied branches such as EEE/ECE)
|
| Preferred discipline of Student
|
| Computer Science and Engineering/Computer Engineering |
| Computing: Collaborative and Social Computing, Computing Education, Computer Systems,Human Computer Interaction |
| Computing: Computer Science, Computer System Security, Software Engineering, Cyber Security & Cyber Physical Systems |
| Information and Communications Technology |
|