Distributed deep learning for dynamic optimisation of constellation resources

About this project

Dynamic optimization of satellite constellation resources plays vital roles for satellite services and their qualities of services but is challenged by large numbers of requests, changing user requirements, unexpected events leading to uncertain request patterns and resource availabilities. On the other hand, dynamic optimization strategies must be scalable but reliable to facilitate on­board deployments and decentralized operations. This project aims to transform user requests of constellation resources as graph­theoretic learning problems where large undirected graphs will be derived to model inter­relations of requests, ground stations across different geographical areas and satellite constellations. Novel graph clustering algorithms will be designed afterward to group requests onto corresponding ground stations while graph­ based multivariate time­series forecasting techniques will be devised to forecast future requests of each node/ground station to improve dynamic optimization performances. The dynamic optimization of constellation resources will be designed both in the centralized (single agent) and in the decentralized manners (multi­agent) based on the deep learning framework using graph clustering information as well as future requests predictions per node. Coordination across each node will need to be designed for the distributed optimization framework to allow improved optimization performances taking advantages of the strength of each agent.

What you’ll do

This project develops the dynamic AI­based optimizations of satellite constellations which timely adapts to changing environments and operational uncertainties. This leads to three research questions to be answered in this project. 

• How to reduce complexity and improve efficiency of dynamic resource optimisation with spatio­temporal clustering of geographically distributed user stations and changing user requests? 
• How to optimise dynamic resource allocations in satellite constellations with Dynamic Deep Q Networks? 
• How to distribute resource optimisation in satellite constellations with Distributed Dynamic Deep Q Networks? 

State­ of ­the art AI methods will be designed here to solve short­ term and long­ term time­series forecasting problems, graph clustering problems and dynamic optimization problems including in the distributed settings where the main goal is to optimally allocate resources of satellite resources.

Where you’ll be based

This project will be hosted by the industrial AI research centre, University of South Australia offering excellent research environments. The Industrial AI research Centre consists of 22 researchers and 14 PhD students. The Research Centre is recognized for its research in artificial intelligence, ontology engineering, integrative private ontologies, knowledge representation, and natural language processing and understanding. It comprises two research groups: the Australian OIIE Interoperability Laboratory and CIAM  Industrial and Applied Mathematics. This project is carried out under the smartsat CRC project titled "SCARLET­: Spacecraft Autonomy & Onboard AI" with AirBus as an industry partner.

Ass/Prof Mahardhika Pratama (primary supervisor) is an associate research professor ­level enterprise fellow in AI at STEM, UniSA. He has been an active researcher with over 100 research papers in top conferences and journals. He won various competitive research awards: top 2% researchers by Stanford University, IEEE TFS prestigious publication award, etc. He is an associate editor in seven prestigious journals and a program committee member in top conferences. He led four special issues in top journals, a workshop in ICDM. As a mid­career researcher, he has successfully supervised to timely completions 7 PhD students and 6 postdoctoral research fellows, his works have been highly cited (h­index: 32, over 3200 citations) and has accumulated over $ 3 million research funding as a chief ­investigator. 

Prof. Ryszard Kowalczyk (co-­supervisor) is the new UniSA SmartSat Professorial Chair in Artificial Intelligence (AI) as well as policy advisor for government bodies in Australia and abroad. He has established track record in AI research with over 250 publications and 4 patents. He has secured over $25 millions in research funding. His research has strong industry engagement: Wipro, DST Group, Zimbani, etc. He holds a lifetime title of State Professor awarded by the President of Poland. 

Dr. Jimmy Cao is a research-intensive academic in STEM (Information Technology) as an ARC DECRA Fellow | UniSA Enterprise Fellow, Senior Lecturer in Artificial Intelligence (AI). His research innovations build the bridge between human cognition and AI and its applications to support decision making. He is also an ACM Distinguished Speaker and taught undergraduate and postgraduate courses in programming in Python, human ­computer interaction, and data analytics. Financial Support 

This project is funded for reasonable research expenses.  Additionally, a living allowance scholarship of $32,500 per annum is available to eligible applicants. Australian Aboriginal and/or Torres Strait Islander applicants will be eligible to receive an increased stipend rate of $46,653 per annum (2023 rates). A fee-offset or waiver for the standard term of the program is also included. For full terms and benefits of the scholarship please refer to our scholarship information for domestic students or international students.

Eligibility and Selection 

This project is open to application from both domestic and international applicants.

Applicants must meet the eligibility criteria for entrance into a PhD. All applications that meet the eligibility and selection criteria will be considered for this project. A merit selection process will be used to determine the successful candidate.

The successful applicant is expected to study full-time and to be based at our Mawson Lakes Campus in the north of Adelaide. Note that international students on a student visa will need to study full-time.

Essential Dates 

Applicants are expected to start in a timely fashion upon receipt of an offer.  Extended deferral periods are not available. Applications close on Monday, 5th of June.