A comparative study of generalised solution concepts for elliptic partial differential equations using nonsmooth analysis techniques. The solution of ellpitic partial differential equations is central to science and engineering. There are a number of solution concepts, such as those of weak solutions and viscosity solutions, but the relations between these are incompletely understood. We shall investigate this major question using recent advances in optimisation theory and nonsmooth analysis. ....A comparative study of generalised solution concepts for elliptic partial differential equations using nonsmooth analysis techniques. The solution of ellpitic partial differential equations is central to science and engineering. There are a number of solution concepts, such as those of weak solutions and viscosity solutions, but the relations between these are incompletely understood. We shall investigate this major question using recent advances in optimisation theory and nonsmooth analysis. Our approach is to use various approximations and their associated second-order subdifferentials, each of which implies a generalised solution concept and associated abstract convexity. Particular attention, including computational details, will be given to equations which have very different solutions of one type from those of another.Read moreRead less
A New Optimization Approach for Tensor Extreme Eigenvalue Problems: Modern Techniques
for Multi-relational Data Analysis. Nowadays, we often encounter complex multi-relational data whose objects have interactions among themselves based on different relations. These multi-relational data can be mathematically modelled as tensors. The tensor extreme eigenvalue problem, which is concerned with extracting the most significant qualitative information from multi-relational data, plays a key role in m ....A New Optimization Approach for Tensor Extreme Eigenvalue Problems: Modern Techniques
for Multi-relational Data Analysis. Nowadays, we often encounter complex multi-relational data whose objects have interactions among themselves based on different relations. These multi-relational data can be mathematically modelled as tensors. The tensor extreme eigenvalue problem, which is concerned with extracting the most significant qualitative information from multi-relational data, plays a key role in modern data analysis. This project aims at developing innovative global optimisation frameworks and reliable numerical methods for tensor extreme eigenvalue problems, and applying the proposed methods to solve various practical problems arising from important application areas such as modern data analysis, medical imaging science and signal processing.Read moreRead less
Computer Assisted Research Mathematics and its Applications. The mathematics community will benefit from infusion of new computer-assisted techniques and modalities for research and training post-graduate students, both from my pure research project and through development of an associated research centre. Ultimately, this should also help more school students learn mathematics well and so play a part in addressing Australia's skill shortage. Also, the work on optimization algorithms promises to ....Computer Assisted Research Mathematics and its Applications. The mathematics community will benefit from infusion of new computer-assisted techniques and modalities for research and training post-graduate students, both from my pure research project and through development of an associated research centre. Ultimately, this should also help more school students learn mathematics well and so play a part in addressing Australia's skill shortage. Also, the work on optimization algorithms promises to improve the performance and quality of many practical signal reconstruction methods. These are used by varied Australian industries from telecommunication to mining and by researchers in the digital arts and fields such as astronomy, physics, chemistry, bioscience, geoscience, engineering and medicine.Read moreRead less
New mathematics for multi-extremal optimization and diffusion tensor imaging. This project aims to establish numerically certifiable mathematical theory and methods for semi-algebraic optimisation problems. Numerically certifiable optimisation principles and techniques are vital for the practical use of optimisation technologies because they can be readily implemented by common computer models and algorithms. Yet no such methodologies exist for multi-extremal, semi-algebraic optimisation problem ....New mathematics for multi-extremal optimization and diffusion tensor imaging. This project aims to establish numerically certifiable mathematical theory and methods for semi-algebraic optimisation problems. Numerically certifiable optimisation principles and techniques are vital for the practical use of optimisation technologies because they can be readily implemented by common computer models and algorithms. Yet no such methodologies exist for multi-extremal, semi-algebraic optimisation problems which are common in modern science and medicine. The expected outcomes of this project include enhanced optimisation methods for diffusion tensor imaging, an emerging technology in brain sciences.Read moreRead less
Innovations in sparse optimisation: big data nonsmooth optimisation. This project aims to produce innovative optimisation methods capable of solving a wide range of practical problems that are currently too complex to be solved. Optimisation involving huge data sets is ubiquitous. Sparse optimisation has emerged as a challenging frontier of modern optimisation because it effectively computes an optimal solution with desired low complexity structure so that a resulting solution can be efficiently ....Innovations in sparse optimisation: big data nonsmooth optimisation. This project aims to produce innovative optimisation methods capable of solving a wide range of practical problems that are currently too complex to be solved. Optimisation involving huge data sets is ubiquitous. Sparse optimisation has emerged as a challenging frontier of modern optimisation because it effectively computes an optimal solution with desired low complexity structure so that a resulting solution can be efficiently stored, implemented and utilised, and is robust to the data inexactness. This project aims at developing innovative mathematical techniques and efficient numerical schemes for solving sparse optimisation problems. The intended outcomes will have significant impact on many areas of science, medicine and engineering, where sparse optimisation is used, including cancer radiotherapy optimal planning.Read moreRead less
Stability of Generalised Equations and Variational Systems. This project seeks to advance a new mathematical theory of variational analysis which may lead to applications in optimisation. The emphasis will be on extensions of regularity concepts appropriate for studying stability (the ‘radius of good behaviour’) of solutions to optimisation problems, particularly those of semi-infinite optimisation and programs with equilibrium constraints, when standard assumptions are not satisfied. The expect ....Stability of Generalised Equations and Variational Systems. This project seeks to advance a new mathematical theory of variational analysis which may lead to applications in optimisation. The emphasis will be on extensions of regularity concepts appropriate for studying stability (the ‘radius of good behaviour’) of solutions to optimisation problems, particularly those of semi-infinite optimisation and programs with equilibrium constraints, when standard assumptions are not satisfied. The expected outcomes may have an impact in enhancing the convergence of numerical methods and facilitating the post-optimal analysis of solutions. It may also generate new tools for increasing efficiencies and cost reductions in engineering, logistics, economics, financial systems, and environmental science.Read moreRead less
Structured barrier and penalty functions in infinite dimensional optimisation and analysis. Very large scale tightly-constrained optimisation problems are ubiquitous and include water management, traffic flow, and imaging at telescopes and hospitals. Massively parallel computers can solve such problems and provide physically realisable solution only if subtle design issues are mastered. Resolving such issues is the goal of this project.
Relaxed reflection methods for feasibility and matrix completion problems. The project proposes to further develop the non-linear convergence theory, and to provide problem-specific implementations. Many applied and pure problems require solution of a large set of linear or nonlinear equations (or inequalities). Highly effective, parallelisable methods are based on iterated projection or reflection algorithms which aggregate information about individual equations. The theory is well developed in ....Relaxed reflection methods for feasibility and matrix completion problems. The project proposes to further develop the non-linear convergence theory, and to provide problem-specific implementations. Many applied and pure problems require solution of a large set of linear or nonlinear equations (or inequalities). Highly effective, parallelisable methods are based on iterated projection or reflection algorithms which aggregate information about individual equations. The theory is well developed in the linear case, but does not explain many important applications for which they are often highly successful (eg optical aberration correction, protein reconstruction, tomography, compressed sensing). The project also plans to provide heuristics to help explain why an algorithm performs well on one class of applications but fails on another.Read moreRead less
Optimal discrete-valued control strategies: A new direction in nonlinear optimal control. The field of optimal control is concerned with finding ways to manipulate systems in the best possible manner. The latest research in optimal control focuses primarily on systems in which the input variables are continuous-valued, yet many real-world systems are controlled via discrete input variables that assume values from a finite set - such as "On/Off", "Open/Closed", "Gear 1/2/3". This project will rev ....Optimal discrete-valued control strategies: A new direction in nonlinear optimal control. The field of optimal control is concerned with finding ways to manipulate systems in the best possible manner. The latest research in optimal control focuses primarily on systems in which the input variables are continuous-valued, yet many real-world systems are controlled via discrete input variables that assume values from a finite set - such as "On/Off", "Open/Closed", "Gear 1/2/3". This project will revolutionise the field of optimal control through the development of new theory and computational tools for optimising discrete input variables in constrained nonlinear systems. The new results will be applied to solve critical problems in the areas of shale-gas extraction, chromatography, pipeline transportation, and micro-robots.Read moreRead less
Data-Driven Multistage Robust Optimization—the New Frontier in Optimization. Robust optimisation is a powerful technology for decision-making in uncertain environments. Yet, developing numerically certifiable optimisation principles and data-driven methods that can be readily implemented by common computer algorithms remains an elusive goal for multistage robust optimisation. But it is crucial for the practical use of multistage optimisation. This project aims to develop this novel mathematical ....Data-Driven Multistage Robust Optimization—the New Frontier in Optimization. Robust optimisation is a powerful technology for decision-making in uncertain environments. Yet, developing numerically certifiable optimisation principles and data-driven methods that can be readily implemented by common computer algorithms remains an elusive goal for multistage robust optimisation. But it is crucial for the practical use of multistage optimisation. This project aims to develop this novel mathematical theory and methods by extending the investigators' recent award winning advances, including the von Neumann-prizewinning Lasserre-hierarchy approach. Results will provide a foundation and technologies for making superior decisions in the pervasive presence of big data uncertainty, enhancing data-driven innovation in AustraliaRead moreRead less