There are 2 PhD student positions available at my group at the Department of Electrical and Electronic Engineering at the University of Melbourne. The students must have a solid background in either optimisation (theory or algorithms), automatic control, information theory, or probability/statistics. Please make sure to contact me (farhad.farokhi@unimelb.edu.au) with a copy of your resume and transcripts if your background matches the requirements. Thanks.
Title: Information Leakage in Control and Optimisation of Dynamical Systems
Description: “Change is the only constant in life”. Attributed to Greek philosopher Heraclitus, this is certainly true, across all time scales, for the dynamical systems at the heart of infrastructure. For example, in buildings, energy demand varies throughout the day and over seasons. Optimal control and optimisation of these dynamical systems based on real-time, or current, data and future forecasts can massively improve efficiency of the infrastructure. Integrated building management and automation can realise 5-15% energy savings, which can amount to significant environmental and economical benefits noting that commercial buildings are responsible for 25% of electricity use and 10% of carbon emissions in Australia. The more real-time data and high-quality forecasts of future user demands is available, the better performance of these dynamical systems can be. However, performance must be balanced with information leakage, referring to privacy and security, risk assessment. On one side of the spectrum resides a blind control strategy that does not gather any user information and relies on sensors that only record low-resolution aggregate measurements. In this case, closed-loop performance of dynamical systems, denoting efficiencies gained from control and optimisation of the system by using real-time data, is restricted, but privacy and security risks are also minimal. On the opposite side sits an advanced controller that makes decisions based on detailed user information and high resolution state measurements. This controller optimises closed-loop performance but the gathered data can infringe on privacy rights and undermine security ideals. This demonstrates the trade-off between closed-loop performance and information leakage in dynamical systems. In practice, due to varying risk appetites and external factors, such as enactment of legislations that restrict gathering, storing, and analysing private data without consent, a careful balance between gathering and processing information and optimising closed-loop performance must be achieved. Hence, a comprehensive framework must be developed to evaluate risks of information leakage in dynamical systems and to balance risk of information leakage against closed-loop performance through optimisation-based sensing, estimation, and control methodologies.
