Discovery Early Career Researcher Award - Grant ID: DE200100063
Funder
Australian Research Council
Funding Amount
$394,398.00
Summary
Nonmonotone Algorithms in Operator Splitting, Optimisation and Data Science. This project aims to develop the mathematical foundations for the analysis and development of optimisation algorithms used in data science. Despite their now ubiquitous use, machine learning software packages routinely rely on a number of algorithms from mathematical optimisation which are not properly understood. By moving beyond the traditional realms of Fejér monotone algorithms, this project expects to develop the m ....Nonmonotone Algorithms in Operator Splitting, Optimisation and Data Science. This project aims to develop the mathematical foundations for the analysis and development of optimisation algorithms used in data science. Despite their now ubiquitous use, machine learning software packages routinely rely on a number of algorithms from mathematical optimisation which are not properly understood. By moving beyond the traditional realms of Fejér monotone algorithms, this project expects to develop the mathematical theory required to rigorously justify the use of such algorithms and thereby ensure the integrity of the decision tools they produce. This mathematical framework is also expected to produce new algorithms for optimisation which benefit consumers of data science such as the health-care and cybersecurity sectors.Read moreRead less
Optimisation of piezoelectric metamaterials: Towards robotic stress sensors. This project aims to design new piezoelectric material microstructures that can enhance the measurement of complex local stress states within robotic limbs. The project expects to generate new knowledge of the achievable properties of multi-poled piezoelectric materials and develop computational tools for the analysis and structural optimisation of such materials. The designed microstructures may revolutionise piezoelec ....Optimisation of piezoelectric metamaterials: Towards robotic stress sensors. This project aims to design new piezoelectric material microstructures that can enhance the measurement of complex local stress states within robotic limbs. The project expects to generate new knowledge of the achievable properties of multi-poled piezoelectric materials and develop computational tools for the analysis and structural optimisation of such materials. The designed microstructures may revolutionise piezoelectric sensor technology. Expected outcomes include manufactured proof-of-concept sensors that enable measurement of local stress fields. This should provide significant benefits, such as improved future robot capability and reliability, and research training for next-generation Australian computational mathematicians. Read moreRead less
Distributed Optimisation without Central Coordination. This project will develop the mathematical foundations for discovery and analysis of iterative methods for optimisation problems in distributed computing systems. Most methods in distributed optimisation were not designed for distributed computing, rather they were adapted for purpose post-hoc. By building on recent advances in monotone operator splitting, this project expects to develop a mathematical theory for decentralised optimisation a ....Distributed Optimisation without Central Coordination. This project will develop the mathematical foundations for discovery and analysis of iterative methods for optimisation problems in distributed computing systems. Most methods in distributed optimisation were not designed for distributed computing, rather they were adapted for purpose post-hoc. By building on recent advances in monotone operator splitting, this project expects to develop a mathematical theory for decentralised optimisation algorithms specially designed for distributed systems. The framework is expected to produce a suite of algorithms, each customised to exploit a specific network configuration. The project will provide significant benefits in distributed machine learning applications such as federated learning.Read moreRead less
Real-time scheduling of trains to control peak electricity demand. This project aims to develop new scheduling and control methods that will enable railways to reduce their demand for electricity during peak demand periods, without undue disruption to the timetable.
These new methods and systems will integrate with—and expand the capabilities of—an Australian train control system that is used by railways around the world. This will enable better management of electricity within a region and be ....Real-time scheduling of trains to control peak electricity demand. This project aims to develop new scheduling and control methods that will enable railways to reduce their demand for electricity during peak demand periods, without undue disruption to the timetable.
These new methods and systems will integrate with—and expand the capabilities of—an Australian train control system that is used by railways around the world. This will enable better management of electricity within a region and better use of renewable energy sources, with significant cost savings for railways and the wider community.Read moreRead less
Stochastic majorization--minimization algorithms for data science. The changing nature of acquisition and storage data has made the process of drawing inference infeasible with traditional statistical and machine learning methods. Modern data are often acquired in real time, in an incremental nature, and are often available in too large a volume to process on conventional machinery. The project proposes to study the family of stochastic majorisation-minimisation algorithms for computation of inf ....Stochastic majorization--minimization algorithms for data science. The changing nature of acquisition and storage data has made the process of drawing inference infeasible with traditional statistical and machine learning methods. Modern data are often acquired in real time, in an incremental nature, and are often available in too large a volume to process on conventional machinery. The project proposes to study the family of stochastic majorisation-minimisation algorithms for computation of inferential quantities in an incremental manner. The proposed stochastic algorithms encompass and extend upon a wide variety of current algorithmic frameworks for fitting statistical and machine learning models, and can be used to produce feasible and practical algorithms for complex models, both current and future.
Read moreRead less
Large scale nonsmooth, nonconvex optimisation. This project aims to develop, analyse, test and apply (sub) gradient-based methods for solving large scale nonsmooth, nonconvex optimisation problems. Large scale problems with complex nonconvex objective and/or constraint functions are among the most difficult in optimisation. This project will generate new knowledge in numerical optimisation and machine learning. The use of structures and sparsity of large scale problems will lead to the developme ....Large scale nonsmooth, nonconvex optimisation. This project aims to develop, analyse, test and apply (sub) gradient-based methods for solving large scale nonsmooth, nonconvex optimisation problems. Large scale problems with complex nonconvex objective and/or constraint functions are among the most difficult in optimisation. This project will generate new knowledge in numerical optimisation and machine learning. The use of structures and sparsity of large scale problems will lead to the development of better models, and more accurate and robust methods. The expected outcomes of the project are ready-to-implement and apply numerical methods for solving large-scale, nonsmooth, nonconvex optimisation problems, as well as problems in machine learning and regression analysis.Read moreRead less
Regularisation methods of inverse problems: theory and computation. This project aims to investigate regularisation methods for inverse problems which are ill-posed in the sense that their solutions depend discontinuously on the data. When only noisy data is available, regularisation methods define stable approximate solutions by replacing the original inverse problem with a family of well-posed neighbouring problems monitored by a so-called regularisation parameter. The project expects to devel ....Regularisation methods of inverse problems: theory and computation. This project aims to investigate regularisation methods for inverse problems which are ill-posed in the sense that their solutions depend discontinuously on the data. When only noisy data is available, regularisation methods define stable approximate solutions by replacing the original inverse problem with a family of well-posed neighbouring problems monitored by a so-called regularisation parameter. The project expects to develop purely data-driven rules to choose the regularisation parameter and show how they work in theory, and in practice. It will also develop convex framework, acceleration strategies as well as preconditioning and splitting ideas to design efficient regularisation solvers.Read moreRead less
Discovery Early Career Researcher Award - Grant ID: DE180100923
Funder
Australian Research Council
Funding Amount
$348,575.00
Summary
Efficient second-order optimisation algorithms for learning from big data. This project aims to apply a diverse range of scientific computing techniques to design and implement new, second-order methods that can surpass first-order alternatives in the next generation of optimisation methods for large-scale machine learning (ML). Scalable optimisation methods are now an integral part ML in the presence of “big data”. While the development of efficient first-order methods has grown in the ML comm ....Efficient second-order optimisation algorithms for learning from big data. This project aims to apply a diverse range of scientific computing techniques to design and implement new, second-order methods that can surpass first-order alternatives in the next generation of optimisation methods for large-scale machine learning (ML). Scalable optimisation methods are now an integral part ML in the presence of “big data”. While the development of efficient first-order methods has grown in the ML community, second-order alternatives have largely been ignored. The project expects to facilitate the development of more effective ML algorithms for extraction of knowledge from large data sets.Read moreRead less
Beyond black-box models: interaction in eXplainable Artificial Intelligence. This project addresses a key issue in automated decision making: explaining how a decision was reached by a computer system to its users. Its aim is to progress towards a new generation of explainable decision models, which would match the performance of current black-box systems while at the same time allow for transparency and detailed interpretation of the underlying logic. This project expects to generate new knowl ....Beyond black-box models: interaction in eXplainable Artificial Intelligence. This project addresses a key issue in automated decision making: explaining how a decision was reached by a computer system to its users. Its aim is to progress towards a new generation of explainable decision models, which would match the performance of current black-box systems while at the same time allow for transparency and detailed interpretation of the underlying logic. This project expects to generate new knowledge in modelling interdependencies of decision criteria using recent advances in the theory of capacities. The expected outcomes are sophisticated but tractable models in which mutual dependencies of decision rules and criteria are treated explicitly and can be thoroughly evaluated. Read moreRead less
Bushfire analytics: optimisation of fuel reduction. Bushfires are an integral part of the Australian ecosystem. However, their severity has been worsening rapidly over the past decade. This project aims to develop a principled and scalable methodology for optimising fuel treatment planning to reduce the potential for severe bushfires. This project expects to generate new knowledge in bushfire fuel management using a groundbreaking combination of mathematical modelling techniques and state-of-the ....Bushfire analytics: optimisation of fuel reduction. Bushfires are an integral part of the Australian ecosystem. However, their severity has been worsening rapidly over the past decade. This project aims to develop a principled and scalable methodology for optimising fuel treatment planning to reduce the potential for severe bushfires. This project expects to generate new knowledge in bushfire fuel management using a groundbreaking combination of mathematical modelling techniques and state-of-the-art optimisation methods. The expected outcomes should provide significant benefits to our nation's ability to respond and adapt to the impacts of environmental change on biological systems and urban and rural communities.Read moreRead less