We are pleased to announce 16 PhD scholarships that focus on creating novel AI research and technologies to enable Missions to solve Australia’s greatest challenges. Below is a brief project description of each project.
Please note, applications are only open for the following project:
- Identification of environmental risk factors for antimicrobial resistance using spatio-temporal statistical models
Applications are now closed for the other 15 projects.
To find out more or to apply for the open scholarship, visit jobs.csiro.au and filter by Studentships and Scholarships. Please contact the supervisors directly for further information.
FAQ: About the PhD Scholarships
Hyperspectral imaging (HSI) is a powerful technology for food quality assessment and safety. Accompanied by novel machine learning and computer vision strategies, HSI can provide an efficient meat surface monitoring solution to detect contaminations and boost public confidence in the meat processing system, while maintaining fit-for-purpose production speed. This project will explore novel, explainable 3D computer vision approaches adapted to the complex nature of industry-relevant HSI data.
Supervisor details
Associate Professor Jun Zhou, Griffith University, jun.zhou@griffith.edu.au
Dr Vivien (Viv) Rolland, CSIRO, vivien.rolland@csiro.au
There is a pressing need to produce more protein more sustainably to meet the dual challenges of human population growth and environmental degradation. Plant proteins, especially from nitrogen-fixing legumes are seen as a great solution to these existential problems. There is an opportunity to build comprehensive protein profiles for diverse mungbean genotypes to accelerate mungbean improvement. Currently, there is no proteomics database available for mungbean.
This project will develop a pan-genome and pan-proteome of the founder lines of a nested association mapping population and use digital twin and AI-guided approaches to determine optimum strategies that will efficiently stack alleles in the shortest time.
Supervisor details
Professor Michelle Colgrave, CSIRO, michelle.colgrave@csiro.au
Professor Ben Hayes, The University of Queensland, b.hayes@uq.edu.au
There is an increasing consumer demand to produce evidence-backed healthy food ingredients and natural functional food products. To achieve this goal, exploring raw materials low in antinutritive content, fortified in micronutrients and enriched in proteins and peptides associated with health benefits is an essential step.
In this project, we will use a combined approach of computational biology, statistics and mass spectrometry-based proteomics to identify novel peptides with potential bioactive function from agriculturally important crop species.
Supervisor details
James A. Broadbent, CSIRO, james.broadbent@csiro.au
Angéla Juhász, Edith Cowan University, a.juhasz@ecu.edu.au
The recent emergence of Digital Twin methodologies, where a virtual representation of a physical asset or process is used to better understand, dynamically optimise and control a process in real-time, has potential for transformational impact in improving additive manufacturing processes.
This project will aim to create an Intelligent Digital Twin combining physics-based modelling, real-time sensor measurements and state-of-the-art Bayesian Machine Learning techniques.
Supervisor details
Gary Delaney, CSIRO, gary.delaney@data61.csiro.au
Chris Davies, Monash University, Chris.Davies@monash.edu
More than 70 per cent of emerging infectious diseases are transmitted from animals to humans, and their incidence has tripled over the past 50 years. Predicting future spillover risks are increasingly crucial to human health and wellbeing, and economic stability.
This project focuses on developing computationally efficient AI/ML algorithms to leverage large spatiotemporal environmental, human, and animal data sets for the purpose of predicting spillover risk. Additionally, we are creating our own data sets through the development and deployment of a novel next gen surveillance system, which will also feed into this project.
Supervisor details
Dr Maryam Golchin, CSIRO, maryam.golchin@csiro.au
A/Prof Roslyn Hickson James Cook University, roslyn.hickson@jcu.edu.au
In many applications, the detection of rare events is of high value, and the automation of such detections using modern machine learning is a desirable goal and open challenge. In most cases, there is plenty of “normal” data but only a small number of rare events either because those events are highly variable, rare, expensive, or difficult to collect.
The main hypothesis of this project is that a self-supervised model could learn “normal” data and automatically detect “abnormal” event. We will investigate anomaly detections primarily from video and other imaging modalities.
Supervisor details
Prof Olivier Salvado, CSIRO olivier.salvado@csiro.au
Prof Clinton Fookes, Queensland University of Technology c.fookes@qut.edu.au
There has been a lot of research interest in using Artificial Neural Networks (ANNs) to accelerate simulations of physical systems. Most initial studies in this area have considered the ANN as a black box for learning the underlying physics of the target system, and have not directly incorporated the known physical laws and natural symmetries of the system into the architecture of the network. This leads to networks that are large and slow to train, and that ultimately yield suboptimal performance. Since no one yet knows how ANNs truly work, the ability to design ANNs is some way off. A more tangible short-term goal is to substantially modify the training process, bringing it a step closer to a design methodology. We call this hybrid training.
In this project we will test the hypothesis that significantly greater performance (training speed, network size, training data requirements, prediction accuracy, and execution speed) can be achieved through a hybrid training approach.
Supervisor details
Richard Scalzo, CSIRO richard.scalzo@data61.csiro.au
Jonathan Manton, University of Melbourne j.manton@ieee.org[Link will open in a new window]
Many scientific programs rely upon publicly accessible geospatial datasets to augment predictions and support greater understanding of nature. Which data to use typically involves complex decision pathways with multiple outcomes possible from a single starting point. Quantitative assessment of data availability, quality and attributes combines to produce a ‘data-richness’ representation that will address an important gap in decision workflows.
This project will produce a workflow to provide national data richness assessments. These assessments will be domain specific, as the relevance of different datasets to scientific question will vary from geoscience, ecology, hydrology, agriculture, marine and atmospheric sciences.
Contact Details
Dr. Mark Lindsay[Link will open in a new window][Link will open in a new window], CSIRO Mineral Resources
/Prof Alan Aitken[Link will open in a new window][Link will open in a new window], The University of Western Australia
While there has been a rapid development of vision-based systems providing accurate and granular quantification of patients’ vital signs, these methods still require a significant advancement to be able to cope with natural conditions such as facial expressions, illumination changes and spontaneous movements. To overcome these issues, we propose to estimate the patient's vital signs directly from a consumer-grade camera using spatio-temporal analysis of video streams and reconstructed 3D models of the patient's face and body.
In this project, the student will investigate and develop novel AI-based methods to deal with the shortcomings of measuring vital signs such as heart rate and respiratory rate from video frames captured in a real-world environment before and during a telehealth visit. Aspirational aims include addressing the feasibility of measuring blood pressure and oxygen saturation using body texture and motion.
Contact Details
Mohammad Ali Armin, CSIRO, ali.armin@data61.csiro.au
Associate Professor Hanna Suominen, ANU, hanna.suominen@anu.edu.au
Long term forecasting of material demand pertaining to global-scale changes like the energy transition, or nascent industries like hydrogen, robotics, or space are currently extremely simplistic.
The guiding hypothesis of this work is that AI approaches including ML and more advanced approaches to parse both structured and unstructured data to provide the basis for better prognostication of technology trends and the implications to material flows at a global scale.
Supervisor details
Tim Baynes, CSIRO, tim.baynes@csiro.au
Prof. Martie-Louise Verreynne, University of Queensland, m.verreynne@uq.edu.au
Antimicrobial resistance (AMR) poses a global threat. The least-well understood component of the one-health perspective to AMR is the role that the environmental component plays in eventual clinical presentations of AMR.
This project proposes a statistical approach that will be applied to a specific freshwater case, but in a way that can be readily generalised to other marine or freshwater applications or case studies. In particular, the project will identify generalisable (surrogate) AMR markers and use them as response variable(s) in a spatio-temporal model that can identify the effect of environmental covariates using observations with point (such as water quality) or areal (such as land use) characteristics.
Contact Details
Keith Hayes, CSIRO, keith.hayes@csiro.au
Erica Donner, University of South Australia, erica.donner@unisa.edu.au
3D scene modelling is a classical problem in computer vision with numerous applications, including robotic mapping, city planning, earth observations, and motion capture. With the recent technological breakthroughs, 3D acquisition methods are now affordable for a broad range of applications. It opens impactful and challenging problems on how to process and efficiently reconstruct 3D scenes.
This project will address the issue of providing fast and affordable 3D scene representations from 3D laser scanning signals and 360 HD images. During the course of this project, the student will investigate recent learning-based 3D reconstruction methods to produce 3D maps of an unknown environment.
Supervisor details
Dr Leo Lebrat, CSIRO, Leo.Lebrat@csiro.au
Prof Clinton Fookes, Queensland University of Technology, c.fookes@qut.edu.au
Despite being mainstream, many consumers will not opt for plant-based milk, in part because of what is deemed inferior taste and texture. Considerable progress has been made towards understanding the sensory and flavour chemistry/ biochemistry of traditional milk and milk products, but this accumulated knowledge is tied closely with providing key input on challenges involved in production of high-quality milk and milk products at scale consistently.
In this project, the PhD candidate supported by a team of multi-disciplinary scientists spanning food chemistry, sensory science, analytical chemistry, -omics and health and nutrition, will deconstruct the classical dairy drinking experience.
Supervisor details
Joanna M. Gambetta, CSIRO, joanna.gambetta@csiro.au
Russell Keast, Deakin University, russell.keast@deakin@edu.au
Wound care in Australian residential aged care facilities has several challenges. The care quality depends on staff experience requiring health specialists, resulting in resident mobility. To avoid the risk and costs associated with resident mobility, telehealth solutions can be adopted. However, the current telehealth solutions for wound care follow a time-consuming off-line assessment approach leading to a lack of timely and coordinated care that reduces its effectiveness.
To overcome these issues, we propose to develop an advanced 3D reconstruction and analysis technology of wounds from images taken with a consumer-grade camera. This new technology will enable a breakthrough in this area by providing objective and precise clinical data to health professionals allowing them to provide more effective care through a telehealth apparatus.
Supervisor details
Dr Rodrigo Santa Cruz, CSIRO, rodrigo.santacruz@csiro.au
Prof Clinton Fookes, Queensland University of Technology, c.fookes@qut.edu.au
Deep learning algorithms have achieved great successes in a wide variety of applications and have been widely deployed in many daily life sectors. Deep models require a large amount of data to train and update. However, it is expensive, time-consuming, or even infeasible to accurately label large-scale datasets. To reduce the annotation cost, cheap methods such as crowdsourcing labelling and automated annotation methods have been widely used. Although cheap data with noisy labels has now been widely exploited to train deep models, little study has been done on selecting reliable models using noisy data.
In this project, we will study model validation using noisy data by modelling the label noise, using the relationship between the noisy and clean validation data.
Supervisor details
Dr Dadong Wang, CSIRO, Dadong.Wang@csiro.au
Dr Tongliang Liu, The University of Sydney, tongliang.liu@sydney.edu.au
Capabilities in effective real-time geospatial data analytics will empower business and organizations to identify markets and revenue sources, deliver timely and tailored marketing campaigns, and make informed decisions in business planning. Despite the enormous benefits of geospatial data analytics, the large volume, multi-modality, dynamic updates, and the timely requirements of analytics tasks all pose great challenges.
In this project, we will investigate two fundamental tasks over geospatial data, large-scale trajectory data analytics and path-based query processing. Firstly, representation learning for privacy preserving trajectory analytics at scale. And secondly efficient path-based query over multi-modal geospatial data.
Supervisor details
Dr Wei Ni, CSIRO wei.ni@csiro.au
Professor Wenjie Zhang, University of New South Wales, wenjie.zhang@unsw.edu.au