Here are the 2018 Project proposals (this list is likely to change). Successful students will be asked to provide project preferences in October.
|Field of Research||Recommended Student Background|
Quantum Combinatorial Optimisation
Professor Jingbo Wang
Combinatorial optimisation is the task of searching for an optimal solution through an extremely large discrete set. Problems in this area are often computationally hard (for example, Max-SAT, travelling salesman, and knapsack problem). This project involves testing a recently proposed quantum optimisation scheme for finding approximate solutions to combinatorial optimisation problems. The aim of this project is to establish quantum supremacy over existing classical algorithms.
|Programming in Fortran, C or Python|
Basic knowledge of quantum mechanics
Strong mathematical skills especially in linear algebra
3D Parallel Large Scale Constrained Joint Inversion of Geophysical Data
Geophysical inversion consists of the inference of the Earth’s subsurface properties such as density, acoustic wave velocity, resistivity, etc. from surface and/or downhole geophysical measurements, addressing Earth as a physical system. It can be used collaboratively with, for instance, probabilistic geological modeling and a statistical description of the petrophysical properties of the studied area.
Integrated geophysical inversion combines geophysical and geological data to allow for a better understanding of three-dimensional geological history and architecture. Integrated inverse modelling relies on three main types of inputs: a geological model (geometry of units), a petrophysical dataset (measurement properties of units), and geophysical measurements.
The team led by Prof. Mark Jessell at the Centre for Exploration Targeting (CET) has developed new technologies that furthered the pre-existing toolset. This is achieved by integrating innovative probabilistic geological modelling combine with statistical descriptions of rock properties to constrain multi-physics geophysical inverse modelling.
This type of approach is computationally expensive as it requires to solve non-linear problems, invert large matrices and process complex multi-modal prior distributions. It may also require to handle and visualize large volumes of data (modelled areas can cover thousands of square kilometers).
Probabilistic models are obtained via Monte Carlo simulation with the aim to adequately propagate structural input uncertainties. This involves heavy pre- and post-processing steps which include costly multivariate analysis. Geophysical inversions are performed using the inversion platform Tomofast-X and requires usage of a supercomputer.
These techniques are part of the toolkit that will be used and built upon by an industry & government-sponsored Cooperative Research Center (the MinEx CRC) and through an Australian Research Council Linkage.
As part of the effort to address problems that will be tackled by these consortia in relation to geophysical modelling, we invite technically-minded students to apply for an internship to work with us at CET.
The aim of this project is to train a student, in collaboration with the Pawsey Supercomputing Centre, to run the codes in parallel 1) to better understand the performances of the inversion platform 2) to perform uncertainty analyses and 3) to visualize large datasets to produce a regional map of anomalies.
The area being investigated is located in the Yerrida Basin (Western Australia), to identify possible prospects for further mineral exploration. This project fits in studies being currently conducted at the Centre for Exploration Targeting; the student will be given the opportunity to contribute to publications and conference work.
Successful students might also be considered to pursue their studies in our group.
Physics/computational physics, Geophysics, Computer sciences, mathematics and/or geosciences.
Character: Interested in tackling challenging tasks and open to multi-disciplinary thinking, capable of working semi-autonomously.
Analysis of some thermodynamic properties of the fluid/fluid/rock interface(s) of a water/h2s/quartz system using molecular dynamics
The presence of acid gases (CO2, H2S) creates problems during hydrocarbon production , and anthropogenic CO2 emissions contribute significantly to global warming . Thus geosequestration has been identified as a key technology to combat global warming, and also to sequester H2S . A vital question that abounds is how these gases, when sequestered, will behave over time in the subsurface. CO2 has been widely studied [4-10] hence the tilt towards trying to understand how H2S behaves in formation fluids. To this end, wettability is a primary factor which determines micrometer-scale multiphase flow. Wettability is also strongly related to the different subsurface gas trapping mechanisms that affect fluid sequestration . Since H2S is a very toxic gas, very little data exist especially at higher temperatures and pressures. Simulation will therefore attempt to compare to empirical data at the lower temperatures and pressures and further investigate higher pressures to reveal the multiphase flow behaviour at a molecular level. Some of the relevant properties that will be looked at include density, interfacial tension and interfacial adsorption of H2S onto water and contact angle of these fluids on quartz surface. Data from simulations will be compared with empirical data and analysed for future studies.
|Petroleum Engineering||Chemical Engineering|
Petroleum/Reservoir Engineering student
Some knowledge of Linux
Good programming background
Simulating Ocean Waves using Swash for Real-time Wave Prediction
In the summer of 2019, Carnegie Clean Energy, in partnership with the University of Western Australia, will be deploying a wave energy converter off the Albany coast. This device will be used to convert wave energy into electricity and will be directly connected to the Albany electricity network. An array of closely spaced buoys will be deployed around the device with the intention of accurately predicting the waves approaching the device in real time. The exact way in which these buoys should be placed and the way they can be used to predict these waves is an important and new field of research.
One of the general design principles of wave energy converters is that their resonance frequency should be similar to the frequency of incoming waves. For maximum power generation, control strategies can be used to tune the resonance frequency of the device based on the (constantly changing) length and height of the immediate oncoming wave. In contrast, traditional ocean wave measurement/analysis has focused on predicting the likelihood and nature of a large wave that might occur in the given sea. These records are useful for the development of ocean ships/structures as they predict the maximum waves the ship/structure needs to withstand, however, this traditional way of observing the ocean is insufficient for maximizing power extraction from wave energy converters. Merely knowing the likelihood of a large wave is insufficient for tuning a wave energy converter’s resonance frequency! In addition, the traditional wave dataset does not sample enough points in close proximity to accurately predict the waves as they move in time and space; nor does it contain enough measurement points for any prediction algorithm to be validated. Therefore a new approach to ocean wave measurement/prediction is needed for optimal wave energy conversion.
In this project, a segment of the coast of Albany will be simulated on the Pawsey supercomputer system using the FORTRAN open source coastal wave simulation software ‘SWASH’. SWASH is a new and novel code for simulating coastal processes. It has been proven to be highly accurate and efficient for simulating coastal waves, however, supercomputer resources are needed to simulate the wide range of wave conditions found around the Albany site.
The SWASH simulations will provide the virtual data needed to determine how the buoys at the Albany site should be placed, and, how effectively they can be used to predict waves. The project will require the intern to: (i) compile and run SWASH on the Pawsey system with the simplified case of waves travelling in one direction over a flat seabed, (ii) use LIDAR surveys to generate a virtual bathymetry of the Albany site, (iii) use this bathymetry to simulate the waves around the Albany coast, and (iv) extract virtual data from these simulations that can be used to simulate an array of buoys and test their wave prediction capabilities. The intern will also have the possibility of extending their work to determine the optimal buoy number and spacing through mass data processing.
|Wave Energy||The student should have a strong background in programming, a passion for simulating real world processes, and basic skills in data analysis.|
A basic understanding of fluid mechanics and ocean waves is also desirable, but no essential.
This position would be ideal for a computer science, engineering, or applied mathematics student.
Analysis and visualisation of flow around different species of coral
Whilst the seabed is often thought of as consisting of sand, much of the worlds nearshore coasts are populated by seagrasses, coral reefs, rocky reefs and other canopy type environments. These environments affect the flow of water by redirecting the flow, absorbing energy or causing the flow to locally slow down or speed up. For coral reefs, these effects on the flow can affect the suspension and transport of sediment, change how nutrients are delivered to the reef communities and affect the shape of the coastline. Yet despite these impacts, little is known about how flow and coral interact.
The aim of this project is to use Computational Fluid Dynamics (OpenFoam) to analyse how flow is affected by the presence of coral. Using 3D scans of coral, laboratory experiments have been previously undertaken to obtain measurements around different types of coral. This project will use the same 3D scans to model flow around the coral in higher resolution to better understand the variability in the flow (in both space and time). Previous studies indicate that this variability can be substantial. The final aspect of this project will be to develop an approach to visualize the 4D results (x,y,z,time) on a 2D page. This project will be integrated into a wider research effort currently being undertaken to understand reefs and how they can be protected as well as rehabilitated.
|Physical Oceanography||Participants are expected to be interested in fluid science and computing simulations. While experience in Computational Fluid Dynamics, and OpenFoam in particular, would be beneficial this is not a requirement. Some knowledge in Linux and MatLab/bash scripting would be helpful.|
pyROM - Python based framework for model reduction
Model reduction techniques reduce the overall complexity of dynamic systems and allow to speed up simulations of their behavior several orders of magnitude while retaining good accuracy. Despite being useful to obtain real-time simulations and apply control strategies, only few freely available software implementations of model reduction techniques have been reported in the literature. Furthermore, the use of these tools tends to be only for a limited range of dynamic problems, mostly related to fluid flows, and to deal with relatively small systems and datasets. Researchers from Curtin University (group of prof. Victor Calo) in collaboration with American University of Sharjah developed pyROM, Python framework for model order reduction. The main goal of the framework is to satisfy the needs of wide range of users to deploy model reduction with good accuracy while still achieving significant computational savings. The project would be ideal for students passionate about learning, improving, and finding applicability of such cutting edge mathematical models in fields such as (but not limited to) - computational fluid dynamics, electronics, computational finance, and solid-state physics.
|Background in mathematics/statistics, Chemical engineering or a related field.|
AutoML vs. Human ML
With past success from various machine learning projects -
i) Desert Fireball Network’s transient object detection (Image)
ii) Ballet movement classification (Sensors)
iii) Aerial rooftop segmentation (Image)
iv) Anomaly detection in escalators (Sensors)
v) Specie classification from camera traps (Image)
vi) Object detection - Identifying craters on martian surface (Image)
Curtin Inst. of Computation (CIC) and Pawsey are now interested in doing a cost-benefit analysis on comparing automated machine learning architectures and human engineered machine learning architectures for various classes of machine-learning problems. Results and benchmarks would be publicly shared. The whole exercise would be converted to an advanced level machine learning course that would be offered jointly by CIC and Pawsey in the near future.
|Machine Learning||Background in electrical engineering, computer science or a related field|
Economic Analysis in the age of Big Data
In data analytics, an accepted framework to explore the relationship between two variables is the regression analysis. Researchers use a sample of data and estimate a coefficient that potentially represent the magnitude of the relationship between two variables. Then to prove the existence of the relationship out of the sample of data (i.e. real-world), researchers evaluate a null hypothesis that assumes the estimated coefficient could be statistically zero. That is to say, what is the probability that the estimated coefficient is zero. And if the null hypothesis could be rejected statistically, researcher would label the estimated coefficient as “statistically significant”. Such approach is useful when the sample of data is representative of the research population, but with the large number of data available from public sources maintaining the same approach seems questionable. That is because the “sample” is so large that, statistically, can be referred to as the population. Such fact challenges the application of null hypothesis evaluation and some of relevant diagnostics tests that are common in the econometrics and data analytics.
Within economics and finance literature there are numerous scholarly works in which authors have utilized a relatively large sample of data and some econometrics method to investigate a relationship. And as long as the estimated coefficients are statistically significant at 5% (that is when there is just 5% probably that the coefficient could be zero), they label the coefficient as statistically significant and take the estimated coefficient as the proof the existence of the relationship. However, because the sample size is large it is possible that 5% significance corresponds to the relationship between the variable and some possible noise in the data, which then makes the reporting of the 5% significance level is arbitrary and informative.
In this project, through several simulation exercises we are to find the threshold where the conventional notion of the null hypothesis testing fails. In other words, we are to find the threshold that any larger sample will bring about significance level, regardless of existence or pattern of underlying relationship.
|Data Analytics background with coding skills in R or Python|
|2018_proj034||Analysis of image processing tools|
|In this role you’ll get familiar with 3D scientific images and conduct a hands-on research analysis comparing the outcome of different image processing tools. The comparison is done in different open-source and commercial tools using various techniques. These techniques include but not limited to image noise reduction, filtering, segmentation, and physical properties calculation. The results will be comprehensively analysed to identify the strengths and weaknesses of different tools.||Scientific Visualisation||Computer science student with strong programming background is preferred. Familiarity with the high-resolution 3D micro-CT images and open-source analysis tools such as ImageJ is highly desirable.|
Glitch Detection in Gravitational Wave Data
One of the biggest problems facing the detection of Gravitational Waves are random "Glitches" in the signal. These glitches saturate the signal and give false detection's of Gravitational Waves. The purpose of this project is to develop an unsupervised machine learning algorithm to detect and classify these "Glitches".
Physics or Maths
Discovering novel genetic signatures of complex diseases through high-performance computing
CSIRO (Floreat, WA)
Complex diseases are influenced by a combination of genetic, environmental, and lifestyle factors, most of which have not yet been identified. A large majority of diseases fall into this category, including Alzheimer disease, asthma, multiple sclerosis, hypertension, Parkinson’s disease and many more. These disorders affect millions of people worldwide (Alzheimer’s disease alone affect more than 40 million individuals), resulting in tremendous public health impacts.
The goal of this project is to develop a novel computational approach to improve the detection of genetic markers associated with complex diseases. This project will take advantage of the parallel architecture of Pawsey’s supercomputer to test a very large number of genetic combinations in order to detect novel associations with Alzheimer’s disease.
This tool will enable scientists to identify biological pathways underlying these diseases which can help with the development of treatments.
Single nucleotide polymorphisms, or SNPs, are the most common type of genetic variation within species. A SNP represents a single letter change in the DNA code. There are about 10 millions of these small genetic variations located throughout our genome, and these vary from one person to another. Collectively, these variations constitute a genetic makeup which is unique to each individual (genotype) and results in a set a unique physical traits (phenotype).
Genome-wide association studies (GWAS) consist in identifying the genetic variations correlated with a specific trait. While some genetic diseases manifest at birth, many late-onset disorders, such as Alzheimer’s disease, do not show any symptoms until much later in life. Therefore, the identification of genetic markers is critical to anticipate the occurrence of such disorders and develop more efficient clinical treatments.
Most currently existing GWAS approaches assume that SNPs confer disease risk independent of other SNPs. Therefore, they are extremely efficient for traits associated to a unique or a few SNPs but cannot be used to detect groups of SNPs which together, lead to a specific phenotype. As many complex diseases result from the interaction between multiple genes, more sophisticated analyses are required. Through this project, we are aiming at developing an approach that consists in testing vast number of genetic combinations in order to identify novel multi-gene signatures that will facilitate the early diagnosis of complex diseases.
|The project is open to any motivated student with an interest in bioinformatics and high-performance computing. |
Knowledge in genetics would be useful but is not required.
Some experience with Linux and Python would be beneficial to achieve results within the ten-week period.
High-Performance Simulation Modelling to Determine Effectiveness of Interventions for Malaria, Dengue and Zika Control
There are two large-scale international programs in international public health, namely Eliminate Malaria and Eliminate Dengue. Both diseases cause large numbers of deaths in tropical regions and agencies such as the Australian Department of Foreign Affairs and Trade, the UK's Weclome Trust and the Bill and Melinda Gates Foundation are coordinating research in this field. Our work at UWA is in a small way contributing to this world-wide program.
Previously we have developed highly detailed simulation models (that is, virtual disease spread worlds) for dengue transmission in Australia and Thailand. These inherently spatial models capture individual human and mosquito movement, and the spread of the virus between humans and mosquitoes.Using these models we have determined the long-term, population-wide effectiveness of alternative mosquito control strategies and the worlds first dengue vaccine. For our dengue vaccination study we have used the Pawsey Supercomputing facilities to model the effect of vaccination cohorts of individual over a period of 30 years, with each day being represented at 4 time-points. Significant amounts of data are generated and analysed.
This new project will use detailed data recently obtained by colleagues in the Solomon Islands and use this to examine a range of disease control methods. These data include incidence and location of malaria, dengue and Zika cases. New control methods will be experimented with in this virtual world, conducting experiments that are infeasible "in the field". This area of study is in the emerging area of discrete-event modelling of infectious disease dynamics. The UWA team are the only Australian researchers who have developed and applied spatially-explicit models for insect-spread diseases, and this is an exciting area for future PhD candidates.
It is hoped that visualisation techniques can be used to describe the geographical spread of these diseases, and their limitation using specific control interventions. These visualisations will be an important means to communicate with international policy makers and so contribute to lessening the burden of these diseases.
|infectious disease modelling||We are looking for a skilled C++ coder. Some discrete maths experience would be helpful, as would some practical experience with ArcGIS. An interest in tropical diseases would be good too, but you are not required to have been infected by any of the diseases under investigation! A book by Frank Ryan titled "Virus X" would be a good starting point and I've a copy I can lend out.|