Are you someone who enjoys a challenge? Would you like to earn money while advancing your career? Do you enjoy learning about new areas? Would you like to strengthen your CV? If you have answered yes to these questions, then read on!

We are offering a unique opportunity for a recently graduated undergraduate student to gain valuable experience while advancing your science career.

In this two year position (with the possibility of a third year), you will join an enthusiastic team of scientists located in sunny St Lucia, Brisbane. You will be involved in the development and application of statistical models for discovering key genes in wheat, one of the world’s most important food crops. As a graduate fellow, you will work on a variety of projects which will afford you the opportunity to be co-author on resulting publications. We will also encourage you to attend a domestic or possibly an international conference to present your work. To help you succeed in this position, we will provide you with a mentor who will support you on your journey as a junior scientist.

This is a fantastic opportunity to kick start your science career by working in a national organisation on real world research that is at the interface of statistics, computing, and biology.

Project context

You will be working with our team on a statistical toolkit for the analysis of genetic data from MAGIC populations. MAGIC is a new type of experimental design in plants and CSIRO has begun recording genetic data on the world’s first wheat MAGIC population. This work is of national and international importance. You will be at the cutting edge of methods development for next generation experimental designs.

Project aims

We expect completion of this position will:

• jumpstart your scientific career,
• give you real world experience as a research scientist,
• train you in statistical genetics and as an applied statistician,
• further your skills in statistics, computing, and biology.

Skills Required

• Background in statistics, mathematics, or an equivalent quantitative discipline.
• Some experience in computing.
• Ability to work in a multidisciplinary team,
• Strong commitment to conducting research with real world impact
• Familiarity with the R programming language,
• Course work in biology and/or genetics.
• Familiarity with UNIX and/or scripting languages.

Skills and knowledge the Fellow will acquire

• Training as a statistical geneticist.
• Experience in the publication process.
• Ability to operate as a research scientist.
• Experience in analysing real experimental data.
• Advancement of written and oral skills.

Computer image analysis is a vibrant research field which sits at the intersection of applied mathematics, signal processing and computer science. The CMIS Graduate Fellow scheme presents an ideal structure and opportunity to provide this breath of training and experience for the right applicant. If you are interested and come from an applied maths, science, physics or engineering background, read on….

The detection and analysis of thin lines (of width less than a few pixels) in images is currently handled by a set of ad-hoc methods such as: edge-detection, thinning, barb-removal, gap-joining, etc., which at best are fragile, slow and error-prone. The fundamental problem is that the discretisation due to pixelation is of the same order as the structures to be analysed. We believe that significant advances can be made by exploiting the underlying local continuity, linearity and smoothness of the structures in sample-space and carefully modelling and inverting the sensing process (pixelation).

The scientific aim is to develop and implement a new set of related methods which are novel and out perform existing techniques. We will also develop a set of efficient algorithms for these new methods. These algorithms will be included into LIAR and VoiR, our in-house Image Analysis R&D software framework. The application of these methods to image data-sets from current insect and plant phenomic studies is also a significant aim as the improved accuracy and deeper analysis of vein structures is an important contemporary requirement in these fields.

Project context

The fellowship is under the guidance of Dr Paul Jackway (Brisbane) and Dr Changming Sun (Sydney) who are Principal Research Scientists with a total of over 25 years experience with the CMIS Quantitative Imaging Group.

The fellowship is consists of four related sub-projects following a typical research workflow:

• Theoretical development of new methods for thin structures. There are two parts to the analysis of thin structures, detection and description. Recent work in our group by Sun & Vallotton (2009) is a good starting point for improved detection. Further, since vein networks have a branching tree-like structure we will implement methods for the concise description and analysis of such networks. Improved methods will be compared and contrasted with existing methods in our libraries and in the literature.
• Fast algorithms for computation of thin structure methods. Various algorithmic “tricks” such as steerability and dimensional decomposition will be fully exploited to maximise speed and efficiency. Emphasis will also be placed on writing code (including documentation) to a professional standard commensurate with our LIAR imaging library.

The fellowship is completed with two application projects which align well with existing major CSIRO projects in automated species discovery and automated plant phenomics:

• The analysis of veins in insect wings. Wing veination is already widely used as a useful taxonomic character for insects. We would hope to improve the accuracy and speed of existing methods.
• The analysis of leaf veins in plants. The application here will attract a new large audience in plant phenomics which will lead to additional publications and uptake of the methods.

Project aims

• This research directly addresses a weakness in the existing image analysis methods for the detection and analysis of thin structures. Furthermore there is a “pull” for such methods from application areas such as phenomics, neurite analysis, hairs, cracks, micro-tubules, and many others in biology, technology, and engineering.
• From a research perspective this project will contribute to knowledge in image analysis and pattern recognition and will be “on-topic” and of ready interest to journals and conferences in the field.
• By enabling the accurate measurement of veins, we will contribute to output projects in automated phenomics of insects and plants.
• This project will add to the capabilities embodied in the LIAR software library and the VoiR R&D framework. From there, if required, the algorithms can easily be ported into, for example, HCA Vision, or other commercial packages.

Skills required

• You will need a degree in applied maths, science, computer science, physics or engineering (or equivalent).
• Some computer image analysis training or experience is preferred. You must love to express your ideas in computer code, any particular language is not so important (“C” preferred), but proficiency is a must.
• willing to learn, want to contribute to globally important scientific research, and get enjoyment from doing this.
• good oral and written presentation skills.
• You must work with integrity, and have a professional attitude to work.
• You must also fit in well in a friendly collegial environment where it is fun to come to work.

Skills and knowledge the Fellow will acquire

• Training, mentoring, skills, and knowledge development in the three components of Image Analysis research: applied mathematics, signal processing and computer science.
• Training in research methods, using the literature, writing and presentation skills in a supportive and encouraging environment.
• Experience in the life-cycle of research from: idea generation, methods development, implementation in software, application to real problems, and publication.
• Attendance at a major international conference (e.g. The IEEE International Conference on Image Processing, Brussels, September 2011.)
• This fellowship will provide a significant incentive to future Ph.D. studies.

Spatio-temporal data often arise from environmental studies for monitoring, assessment and natural resource management. These data typically are generated by measuring multiple variables (or indicators) at a number of locations in a region (which may provide information at a variety of spatial scales from local to state-wide) and points in time. A key objective of analysing such spatio-temporal data is to develop meaningful and informative statistical models that may be used for descriptive and/or predictive purposes. Two challenges in meeting this objective are selection of appropriate predictor or explanatory variables and incorporation of an appropriate (sptaio-temporal) correlation pattern.

The Graduate Fellow (GF) will help explore and develop spatio-temporal methodologies where the focus is on modelling the trend, variance and correlation structures of data from various sources and with differing spatio-temporal scales. To enable the GF to experience a variety of environmental application areas, we propose the four specific applied projects to motivate innovative research directions.

Briefly these are

1. Water quality monitoring data: the spatio-temporal water quality data exhibit extreme skewness and data censorship so model development needs to account for all of these features;
2. Fisheries data (in collaboration with Wealth from Oceans and CSIRO Marine and Atmospheric Research scientists): often exhibit many zeros and are spatially and temporally correlated. There may be more than one species of fish/prawn to consider, in isolation or jointly. Explanatory variables could be used to test the degree of zero-inflation. Model selection will play an important role in developing an appropriate model for such correlated zero-inflated count data.
3. Pollutant loads in rivers: model selection will play an important role for developing a valid predictive model but also for studying the efficiency of different monitoring designs. Applications to data from Queensland rivers and catchments that directly impact on the Great Barrier Reef health, so highly relevant for developing policy initiatives and water quality improvement plans.
4. Ecosystem health data (in collaboration with QDERM): this data comprises multiple ecosystem health indicators measured at various locations across a large-scale domain. The motivation for collecting the data is to form an assessment of the health of the ecosystem for that domain and to make inferences about risks to the condition of that ecosystem. Analysis of this data is not straightforward for a variety of reasons including design and practical considerations that need to be taken into account. Complex systems such as river networks often demand complex models so model selection will be critical to ensure.

We believe the novel applications of newly-developed statistical tools to these important environmental projects will lead to joint publications in high-impact international journals.

Project aims

The main responsibilities for the GF include: managing the datasets; understanding the background to the problems (e.g. perhaps by preparing a review on the subject); contributing to team discussion; undertaking data analysis under close supervision; carrying out simulation studies for understanding model sensitivity and robustness; documenting findings from these studies through writing technical reports; presenting key findings at relevant meetings/conferences; and contributing to writing up key findings for journal articles.

Skills required

• 3rd-year level statistical knowledge of topics such as statistical inference, applied statistics including regression analysis, generalised linear modelling, etc., and experimental design and sampling
• Familiarity and basic knowledge of the R statistical programming language

Skills and knowledge the Fellow will acquire

• Good understanding of statistical theory and modelling skills for real world data analysis
• Problem-solving skills
• Background knowledge on some key environmental research areas
• Experience of working in a multidisciplinary team
• Improved computational, scientific writing and presentation skills

The prediction of Y|X is a well studied problem in statistics and data mining. There are a large variety of tools and approaches in the literature. A less studied problem is how to generate and select appropriate covariates. While it is well understood that the appropriate transformation of the X’s can significantly improve prediction performance, approaches to explore this are less common than may be expected.

This project will explore this problem. It will consider the generation of covariates based on potentially incomplete knowledge of the functional form of good predictors suggested by theory and domain knowledge. The approach will consider generating large numbers of plausible covariates and developing approaches to screening them. Examples include:

• In biosecurity/biodiversity applications the prediction of the spatial distribution of organisms is a key consideration. This is typically related to environmental conditions but it is well understood that typical variables such as mean temperature are only weakly related to better predictors related to process such as stress and growth indices. What are the optimal indices is unknown. There is potential to investigate screening across families and classes of stress and growth indices.
• The problem of statistical downscaling of climate forecasts involves the selection of appropriate covariates from a very large set of possible variables, and the modelling of local climate variables in terms of these. The covariates in the statistical model typically come from Global Climate Models and these provide data over a large grid of values. Variable selection methods have been developed for linear regression models in this case, but it is know that certain types of non-linear models are preferable for longer term climate forecasts, and so the issue is how to select appropriate covariates in this case.
• In managing more regulated water ecosystems such as Murray-Darling Basin there is a need to understand how choices that affect the hydrology in turn affect the ecosystem responses (birds, fish, tree health etc) that we value. A key challenge is the identification of those characteristics of stream flow that are predictive of ecosystem health. These characteristics, or covariates, are typically unknown but relate to aspects of the intensity, frequency and duration of flow events. Knowledge about them could then drive informed decision-making about the provision of water for the environment (known as ‘environmental flows’).

An associated challenge is how to generate a robust and meaningful cluster-driven regionalisation of major water ecosystems based on a suite of flow characteristics that best capture long term hydrological records.

This project will review application of p > n techniques in environmental sciences. They will explore issues arising from generating large numbers of “plausible” covariates. Issues such as the trade off between p and n will arise as well as identification problems and developing novel representations of these so they can be objectively considered.

Robustness of sets of predictors could also be an important consideration in the project as a requirement of the method is that small changes in the data do not lead to large changes in the set of predictors selected. For example, the choice should not be sensitive to a few data outliers or errors.

Most variable selection techniques have been developed for linear regression problems and the mean rather than more general regression approaches. There is an opportunity to extend these methods to quantile regression and non-linear models.

Other Project [80% of time]

The person will also be expected to contribute to other projects in conjunction with other Canberra staff. The Canberra lab is a diverse group of staff working in a number of high profile such as Water, Climate Change and Adaptation, Biosecurity, Ecosystem Health. The person would be rotated through a broad range of projects to develop skills and experience. These projects may align closely with the research project, as indicated by the examples above, or they could be in different areas if it was thought that the Fellow would benefit from a broader exposure to statistical problems.

Project aims

The Fellow will be assisted to develop a project in this area. The Canberra lab has researchers working in a range of application areas as well as research expertise in statistics and data mining.

The project could be:

• Computational: looking at issues arising from fitting very large numbers of models and representing the output in low dimensional, interpretable forms. Use of novel visualisation techniques would be a key component. This project would utilise the statistical and data mining expertise in the Canberra lab.
• Empirical assessment of the behaviour of particular approaches, using this to elucidate more general understanding. There are a range of applications being considered in the Canberra lab that could underpin this.
• Analytic exploration of a particular aspect of this problem. The scope would need to be appropriately narrowed. This would include involvement of staff from the Australian National University.

This choice of topic will be matched to the student’s strengths and aspirations.

Other Projects

The person will be expected to:

• Actively contribute statistical thinking and analyses, and communicate / report findings.
• Work in multi-disciplinary teams.
• Show a commitment to developing their statistical skills.

Skills required

• Degree with strong statistical/ computational component / major.
• Experience with a range of regression modelling and variable selection techniques.
• Computing skills (desirable)
• Familiarity with R statistical language or other computational programming language like C/C++ (desirable)
• An enthusiasm to address environmental issues with the innovative use of statistics

Skills and knowledge the Fellow will acquire

• Experience in statistical thinking, and applying it to a range of different application areas.
• Applied statistical skills.
• Development of research thinking
• Team work and collaboration
• Incorporating practical constraints when problem solving
• A broadening of knowledge into the application area chosen
• Communication and scientific paper writing skills
• Experience in conference presentation

The computational prediction of macroscopic properties of materials has wide applicability in many areas of applied science and technology. One area of particular current interest to CSIRO and the oil and natural gas industry is the prediction of various petrophysical properties of rocks. This is important in circumstances where experimental approaches are not practical. Some of the properties of interest include mechanical, electrical conductivity, dielectric, and transport properties such as hydraulic and thermal conductivity. The starting point for this kind of calculation is a digital representation of the microstructure of the rock. This can be based on x-ray absorption ( 3D tomography) data.

There are two components of this research project:

Data regularisation: Typically there is only a restricted number of channels of x-ray absorption (tomography) data available. Strictly speaking, this is not sufficient to derive the kind of detailed (that is, high dimensional) microstructural and compositional data that is often required; however, a simple Bayesian regularisation approach will be investigated to try to resolve this indeterminacy.

Property prediction: This will investigate ideas from multigrid and domain decomposition approaches in finite element analysis to enable the feasible estimation of material properties from these digital representations of material microstructure.

Both these activities are computationally intensive, and are well suited to multiple processor implementations. Parallel algorithm development and implementation will be a large part of this project.

This project will involve collaboration with other researchers in CSIRO Petroleum Research, the CSIRO Wealth from Oceans Flagship and the CSIRO Computational and Simulation Science Transformational Capability Platform.

Project aims

This is a “proof-of-concept” project whose aim is to investigate the two ideas described above. If these ideas prove fruitful, the outcomes of the project will provide the basis for seeking larger scale engagements with both CSIRO and industry groups.

The primary role of the fellow will be in programming and algorithm development (under supervision). However, there will also be non-trivial analysis and model formulation aspects of the work, and so the fellow will also have an exposure to the need for deeper thinking about the underlying mathematical and statistical issues. It is also expected that the fellow will write a number of reports describing the work they have done.

Skills required

• Skills in real analysis and linear algebra.
• Skills in numerical mathematics and experience in using a scientific computing language (e.g. fortran, C, or C++).
• Background knowledge of statistical concepts.
• Background knowledge of mathematical modelling of physical processes.

Skills and Knowledge the Fellow will acquire

Opportunity to obtain experience in computation and algorithm development, including parallel and multiprocessor computations. This area is likely to be a big part of the future in the quantitative sciences.
Exposure to a range of mathematically based methodologies and areas in the applied quantitative sciences, and the opportunity to collaborate with other application scientists.
Experience in writing technical reports and documents.

Biomechanics is an exciting and rapidly evolving field with applications in orthopaedics, vehicle safety, health and human ergonomics. The focus of the proposed project is to develop improved methods in musculoskeletal biomechanics to inform decision making and reduce costs in musculoskeletal healthcare.

The Computational Modelling (CM) group at CSIRO are developing the capability to address evolving challenges in biomechanics by applying their expertise in discrete element modelling (DEM) and smooth particle hydrodynamics (SPH). These techniques have advantages for impact studies, particularly where fluid and discrete particles are involved, such as water sports and wear within orthopaedic implants. Integrating DEM and SPH into a biomechanics framework would provide a competitive capability to address the growing biomedical and biomechanics challenges for Australia’s present and future.

Project Context

This project is important to the broad community in the areas of general public health and safety. The opportunity to inform decision making and reduce public health spending is always a benefit to the Australian community. The methods investigated will be novel and provide an opportunity to make unique contributions to the scientific community. The themes of this project can lead to further postgraduate studies or a PHD project in partnership with the CSIRO.

Project Aims

A two year project is proposed, which will involve novel computational capability development (approx 30%), and several application-based problems where the group’s DEM and SPH techniques have significant advantages (approx 70%).

Year1:

• Methodology: The computational modelling group has developed the capability to model human motion driven by prescribed kinematics. The current modelling software can also predict dynamic musculoskeletal motion. The ability to switch between these modes is both novel and a current modelling challenge. The fellow, lead by Dr Fernandez, will be exploring algorithms to switch between prescribed and predicted motion. This will require skills in control theory and/or multi object optimisation. This approach is entirely new and will aid in modelling sudden impact response in applications such as car crashes and aeroplane turbulence.
• Applied Project 1: Solve a generic musculoskeletal problem of a human interacting with solid objects to study how the subject modifies their posture in order to maintain balance or prepare for impact. Focus the application on scenarios as a car crash or during plane turbulence.
• Applied Project 2: Develop a demonstration showing the response of humans to impact from waves. Focus on scenarios, such as the human response to being hit by a wave on a ship deck and being dragged by a rip at the beach.

Year 2:

• Methodology: The human musculoskeletal system is comprised of complicated 3D muscles with different bone attachments and muscle actuation abilities. Most importantly, muscles are required to drive the bones during dynamic motion. The fellow, lead by Dr Fernandez, will contribute to two components of modelling the 3D muscles. Firstly, the fellow will contribute to building key muscle geometries from MRI and/or the visible human database. Secondly, they will perform a literature review to identify simple lumped parameter models to describe the muscle behaviour. These chosen methods will need to perform efficiently in our software.
• Applied Project 3: Develop a model of a human with full musculature walking into an object to observe how the muscles behave during sudden impact. Compare the human muscle responses with Electromyography (EMG) and the literature to provide evaluation of the models.

Skills Required

• Knowledge of Fluid Dynamics, Physics, Mathematics or Engineering.
• Knowledge of optimisation methods and basic control theory.
• Knowledge of Unix operating systems, computer programming (Fortran 90, c, c++).
• Use of scientific commercial software (eg. Matlab).
• Good written and oral communication skills; proficiency with Word and Excel.

Skills and knowledge the Fellow will acquire

• The student will build on general mathematical skills learned during their undergraduate studies.
• The student will become well rounded in biomechanics methods, computational modelling and gain valuable paper writing skills. These skills will make the student attractive to CSIRO as a potential PhD student or junior research scientist.
• The student will also gain valuable experience using specialised commercial animation software.

Extreme geophysical flows such as dam break flooding, landslides, tsunamis and storm surges are an important class of problem in civil engineering and environmental flows. Their prediction is important to evaluate the risk of damage to surrounding life and infrastructure. In some instances as in dam break flooding such modelling is now a required element in the design of a dam and its surrounding environment. The impact of these flow events on developed areas can be sufficient to completely destroy infrastructure such as roads, railways and bridges and to demolish buildings. Additional features of such extreme flooding include movement of large amounts of sediment (mud) and debris causing secondary landslides along with the risk of distributing pollutants from sources such as chemical works or mine workings in the flood risk area.

Project Aims

A two year project involving a combination of application oriented work delivering into internal and external projects and contribution towards development and testing to evaluate the combined SPH-FVM solver is proposed. Specific project goals over the two to three year period are described below:

• Perform an extensive literature search on the state of the art in tsunami modelling and identify gaps in this area especially in the prediction of tsunami inundation. Prepare a technical report on the literature survey.
• Identify and obtain relevant data such as inundation maps to compare simulations against tsunami events.
• Develop a capability to automatically determine the underlying river-bed topography based on DTMs provided by clients which have the water included in the terrain model.
• Contribute towards simulating various dam break flood scenarios for SMEC, Australia and CASM, China and storm surge scenarios for the WHC Flagship as needed.
• Contribute towards extending the FVM based shallow water solver to include the ability to use unstructured meshes.
• Apply the developed combined SPH-FVM solver to simulate the historical St. Francis Dam Break model all the way to the flood plain.
• Based on the data obtained, contribute towards simulating a tsunami event using the combined SPH-FVM method.
• Contribute towards writing a journal paper on the simulation of the St. Francis Dam Break to the flood plain by applying the combined SPH-FVM method.
• Take the lead in writing a journal paper to illustrate the capability of the combined SPH-FVM method in simulating a tsunami event starting from its origin up to and including coastal inundation.

Skills required

The prospective candidate will need to have skills in the following areas:

• Programming skills in Fortran 90/C/C++
• Good knowledge of Fluid Mechanics
• Good written and oral communication skills
• MS office skills including word and excel
• Knowledge of working in UNIX/LINUX environments.

Skills and knowledge the Fellow will acquire

• The project provides the opportunity for the graduate fellow to appreciate the use of simulation tools for real world applications involving large scale geophysical phenomena. This is especially important in today’s context where occurrences of natural disasters are becoming more common. They will also develop an understanding of the Smoothed Particle Hydrodynamics (SPH) method that has started becoming increasingly popular for solving complex free surface problems in the environmental and fluid/structure interaction application domains.
• The graduate fellow will be involved in cutting edge research in the area of computational flow modelling resulting in a unique capability to solve extremely large geophysical flows very accurately and efficiently using the combined SPH-FVM solver.
• The research carried out during this period will motivate the graduate fellow to pursue a doctoral degree with CSIRO.
• The graduate fellow will be an author in at least 2 journal papers in high impact factor journals.
• The graduate fellow will have the opportunity to attend a major international conference.

This fellowship is designed to support the research needed for the Centrelink Human Services Ecosystem project. The Centrelink work involves a detailed analysis of the complex interactions involved in the delivery of human services across Centrelink. In particular, we seek to provide insightful methods for the assessment and optimal selection of programs, interventions, service delivery footprints and resource allocations across Centrelink, based on changes in external demographic, economic and social factors.

Such a system requires a conceptual framework to capture the complex relationships that underpin the interlinked behaviour of Centrelink’s customers, Centrelink’s service delivery mechanisms, government policies and programs, and the broader social environment, and to develop analysis tools that utilise this framework as the basis for quantitative predictions about behaviours and outcomes. We propose to develop this framework based on the notion of a “Human Services Ecosystem”. This ecosystem will be used to classify the different groups in the system, describe their interactions, and trace their evolution over time in response to changes in the broader socio-economic environment.

The central idea is to describe the Human Services Ecosystem and other complex systems in terms of the short-term and long-term trajectories of customers, or entities, through a variety of processes and facilities. That is, we want to identify the pathways entities can travel, as they move through transactional units in the environment and states in their own trajectory. The benefit from this holistic modelling is expected to include: more accurate analysis, forecasting and modelling of demand for services; enhanced analytical approaches to matching resources to needs; successful integration of complex high volume data (particularly transactional data) and information systems underpinning decision making systems; better understanding of capacity utilisation; and identification of the least cost and efficient opportunities for process and infrastructure innovation and improvement.

Project Aims

The goal of the project is to develop the concept of clinical pathways used in healthcare into a more general modelling construct methodology and apply it to Centrelink and possibly other applications, particularly (but not exclusively) those involving human systems. The idea of stochastic pathways is to try to understand a complex system by focusing on the paths that can be taken by an entity moving through the system. This path would often be stochastic in that the move from one stage or state to the next is not always predictable in advance.

We are interested in developing mathematical tools to see what properties of a system can be determined by analysing it in terms of the pathway descriptions. This should lead both to better understanding of the behaviour of the system and allow the development of optimisation tools for re-designing the system or changing resourcing decisions in order to maximise the performance of the system.

The successful graduate fellow will spend time on methodological research in stochastic pathways modelling and be involved in practical projects related to these particularly in the context of the Centrelink. In addition there are also likely to be opportunities for the fellow to be involved in some other short-term projects in the MOS group in order to broaden their experience. For example, this might include supply chain optimisation or looking at paths of packets through sensor networks taking into account transmission and node failures.

Some of the projects that the graduate fellow is expected to be involved in include:

1. Literature Survey: To provide background on research into stochastic pathway models in different domains.
2. Setting up Sample Centrelink Pathways: To show how the pathway concept can be applied in Centrelink. The student will be involved in some of the data analysis work.
3. Computations using pathways: To investigate whether data anomalies can be identified and allowed for in the Centrelink database. Also to test whether, and how accurately, pathway structures can be automatically extracted from transactional data.
4. Methods for identifying Human Services Pathways: To use statistical, decision technology, behavioural economic and other social research methods to find, explain, and identify methods for influencing behaviour in, pathways.
5. Development of a simulation environment to simulate movement of entities through stochastic environments as test bed for proposed analysis method. Implementation of experimental analysis methods for stochastic pathway models.
6. Methods for improving Human Services delivery using Pathways: To show how various pathway and network analysis methods can be used to forecast the demand for services as the system evolves, to identify the impacts of proposed policy/process changes, and to deal with uncertainty in the system.

The graduate fellow will be working closely with not just their supervisor but also other members of the people oriented systems stream and the MOS group.

Skills required

The Fellow will need to have a general knowledge of Operations Research methods in the area of network flow modelling. Other skills include:

• Knowledge of the use of pathways in health services (desirable).
• Programming in C++, Python (desirable).
• Knowledge of simulation and/or system dynamics modelling (desirable).

Skills and Knowledge the Fellow will acquire

• Network flow modelling, with an emphasis on stochastic flows,
• General database and statistical analysis.
• Simulation and optimisation modelling.
• Preparation of client reports and journal articles.

It is expected that the student will be given the choice between two projects.

1.Real-time BACI designs and analysis:

Sydney Catchment Authority have provided dairy farmers along the banks of supply water with grants to build filtration ponds for washing out milking sheds and grants to build fences for restricting animal access to the rivers. These grants are offered to several farmers in the same catchment. Each project has a different starting and completion date - ranging over several years. The traditional approach to assess impacts of such interventions is to collect data on water quality at both control sites and downstream of the impacted sites before fences and/or building filtration ponds. Then, after all the projects have been completed, collect data on water quality at these same locations again. Thereafter the data are used to carry out the traditional BACI analysis. This approach requires decision makers to wait several years before the significance of the impact can be tested thus delaying further investments in such interventions if they are highly successful. The approach we are proposing is to make use of the routinely collected data to see if a real-time BACI design can be carried out. This will have several before and after indicators, one for each project, and thus try to estimate the impact of each project as well as the accumulative impact of all projects in real-time. This would allow decision makers to gain timely feedback on the significance of the interventions, and provide greater diagnostic information on what interventions have the biggest impact, e.g., fencing versus filtration ponds. A pilot study with the Sydney Catchment Authority is expected to start in the next three months, but we would expect the student to complete the study with a more extensive study involving several interventions and a longer after period for the pilot study.

2.Real-time disease surveillance using all of Australia/Queensland data:

This study depends on us getting data for all of Australia (preferable) or at least most of Queensland. Multivariate spatio-temporal disease surveillance involves monitoring several related diseases (e.g., influenza and pneumonia) jointly trying to determine spatial-, age- and gender-related disease clustered outbreaks. An example of a clustered outbreak is a time period where the counts of both influenza and pneumonia at the Gold Coast is significantly higher than expected for children under 13 years of age. An example why it is important to detect outbreaks early is evident in the recent swine flu outbreak that influenced pregnant women adversely – early detection of such clusters has the potential to save the lives of both the mother and the expectant child. Multivariate monitoring plans are expected to have more power than running separate surveillance plan for each related disease, that is, the exploiting the cluster sources of variation helps detect the disease much earlier.

This research study will:

• a. Develop (multivariate) spatio-temporal models for forecasting one time point ahead for predicting the usual behaviour of disease counts (expected values).
• b. Monitor the departure of actual counts from their expected values to establish unusual increases in disease counts.
• c. Assess the value of multivariate surveillance relative to univariate surveillance in terms of early detection.
• d. Compare and control the false alarm rates of plans, e.g., multivariate versus univariate.

Project Aims

• Develop the person multivariate problem solving skills with a range of applied statistical problems.
• Develop models for establishing expected values for measures (both counts and continuous). Models here will be (multivariate) spatio-temporal models.
• Develop robust and efficient estimating technology for fitting models.
• Develop monitoring technology that has the ability for detecting anomalies early.
• Fitting models to longitudinal/time series data from Centrelink and performing longitudinal/time series data analyses.
• Research project in disease surveillance and/or adverse drug reactions.
• Learn the skills of writing efficient software from our resident software engineer, and learn the skills needed for handling very large datasets.
• Learning to break complex tasks down into (hierarchical) parts, and learn to make reasonable assumptions when dealing with complex applied problems.
• Develop their research skills to the level where they will be comfortable presenting a paper at an international conference.
• To further develop the multivariate spatio-temporal modelling and monitoring capabilities in CSIRO.
• Produce a research paper than can be presented at an international conference, and be published in an international journal.

Inputs expected from Fellow

The fellow should have an interest in the application of statistics and be keen on developing their (statistical) problem solving skills. They should be comfortable operating in a team environment and have the ability to take direction from team members. They should be prepared to study up aspects that are outside their university (and other school) curriculum that are necessary to solve the problem they are working on. They should also be keen on developing into independent thinking and creative scientists.

Skills required

The fellow should have good effective communications skills, and have had some study experience with multivariate analysis, cluster analysis and (times series) modelling. Reasonable programming skills and some exposure to statistical packages such a R or Splus.

Skills and Knowledge the Fellow will acquire

Developmental projects:

• Help in Hydro-Tasmania and CSIRO partnership for developing automatic QA/QC algorithms that can help maintain and monitor both measurements and models.

1. Excellent multivariate spatio-temporal interpolating models are needed for assessing whether measurements are multivariate spatio-temporally consistent with what the model suggests as expected.
2. Efficient updating technology (recursive estimation methods) for updating parameter estimates.
3. Monitoring technology for identifying model and gauge biases. A developmental or methodological project (roughly 20% effort) that the Fellow will be expected to present on at a reputable conference, ideally an international conference, at the end of their second year.

• Sydney Water – estimating deterioration rates in the sewerage pipeline network by estimating what fraction of the total rainfall per storm that gets into the sewerage pipelines in any given catchment, and participating in upgrading of the SWAGMAN software with this capability.
• Sydney Catchment Authority – work on a modelling and real-time (daily) monitoring capability for water quality variables for identifying the risk associated with several key analytes. Conduct real-time before, after, control and impact (BACI) designs for assessing the impact of interventions that are slow to complete (take several years to be completed). This is a potential research project for the graduate fellow. (If this is selected as the project it will be written up as a research paper for presentation at an appropriate national or international conference – potentially ISI 2011).
• RTA QA/QC is needing a significant attention from a CMIS statistician, and the Graduate Fellow will potentially work with Chris Okugami and John Donnelly in helping develop key technology to the RTA in QA/QC and traffic count modelling.
• Disease surveillance – potential partnership with ABIN and/or Queensland Health – the student could be involved in the development of models and monitoring technology for early outbreak detection. This is a potential research project for the graduate fellow. (If this is selected as the project it will be written up as a research paper for presentation at an appropriate national or international conference – potentially at the 2011 Advance Disease Surveillance conference in the USA).
• Centrelink have real-time surveillance and longitudinal data analysis needs that will require substantial effort. The data are observational in nature with the occasional intervention by social workers, and it is likely that the graduate fellow will have some exposure to case-control designs and surveillance methods.
• Training in presentation will be first done by presentations within the friendly environment of the local seminar series and young statisticians’ conferences. Writing skills will be provided by participating in client reports, writing of technical reports, algorithm documentation and at least one research paper.
• John Donnelly is project managing a success project with Right Ship and they are likely to revisit our services in terms of evaluating their vessels risk assessment process as they try to improve the process based on our recent work. This task is likely to revisit some of our evaluation processes, and it is an ideal project for the graduate fellow to learn more about risk assessment and the assessment of risk assessment processes.
• The graduate fellow will be given a research project to analyse the recent influenza outbreak in the Gold Coast (QLD) with the aim to develop an innovative early warning system for disease outbreaks. This will include age, gender, location and critical outcome differentiations with the aim of monitoring the size and severity of the outbreak once detected. This work will be presented at both an international conference and a national conference.
• The student will be exposed to other applied solving projects that come along within the tenure of the positions. The student will be taken through the problem solving task from the beginning to developing a solution, followed by the communication of the results to the client. This process will also include the recommendations for future work. The student will learn about the consulting process, the stages of concern (with associated risk factors), and the need to keep the client informed throughout the project task.
• Michael Buckley, Glenn Stone and David Clifford have projects that may broaden the experiences of the student. Interesting projects that surface will be evaluated on a case by case basis in terms of providing the graduate fellow with an experience that will broaden his/her future capabilities. These with be short term applied research projects where the student could be suitably deployed and provide useful support for the research scientist.
   

Fellows assigned mentors are Louise Ryan (disease surveillance), Ross Sparks, Peter Toscas, Chris Okugami, John Donnelly, Michael Buckley, Glenn Stone and David Clifford.

Fellows will also be offered various opportunities for formal training which include local reading group activities, Australian Statistical Society workshops and AMSI workshops.

3D object reconstruction and analysis are integral to developing cutting edge technologies and applications such as 3D plant modelling/analysis for plant phenomics applications, 3D insect modelling/classification in the area of entomology.

The objective of this proposal is to develop advanced techniques, algorithms, and systems for 3D reconstruction of objects from multiple 2D images or other types of data sources and to develop capabilities to analyse such 3D data. These 3D models can then be used for measurement and characterization of plant and insect.

The significance of this capability for plants is that an accurate 3D model can be used to quantify genetic variation in growth of the whole plant, organs of the plant and the model can be overlayed with other spectral information. Processing thousands of models and extracting quantitative information without human intervention is necessary for high throughput plant phenomics to succeed. Image-based phenomics has the potential to yield deeper insights into structure, development and physiology by capturing information in greater detail and frequency and with greater objectivity than traditional methods.

For insects, the high throughput rendering is equally important, as is the quantification of organ volumes, areas etc for insect identification and classification. Characterising insect images in morphologically meaningful ways would help create globally accessible taxonomic resources with applications in biosecurity, biodiversity and functional genomics.

The 3D reconstruction and analysis system developed for plant and insect will also deliver outputs for other applications. Research in image-based phenomics would help ensure these initiatives fulfil their potential as dynamic international research facilities with particular relevance to Australia.

The fellowship is under the supervision of Dr Changming Sun in the CMIS Quantitative Imaging group.

Project aims

This graduate fellow is a “pathfinder” towards a much larger research effort. The Fellow’s research would stimulate wider support and investment, enabling CSIRO and collaborators to deliver even greater benefit. This project will contribute to knowledge in image analysis and pattern recognition and will contribute to the automated phenomics of plant and insects.

Skills required

• A degree in computer science, physics or engineering (or equivalent).
• Image processing or image analysis course or experience is preferred.
• C or C++ language preferred.
• Good communication or presentation skills.
• Any awards/prizes a plus.
• Any publications a plus.
• A highly motivated, independent and creative graduate.

Skills and knowledge the Fellow will acquire

• Exposure to high-impact practical research challenges, scientific expertise in multiple disciplines, and scientific infrastructure.
• Exposure to technologies and experiences in developing in-demand image-based phenomics methods will lay the ground for an exciting scientific career.
• Professional development through attending international conferences and publishing research results.
• Learning technical skills in image analysis, programming languages, etc.
• Chance of possible future PhD study.

Quantitative imaging, biotech imaging. Given the nature of the methodology we plan to deploy the algorithms on the GPU cluster.

Markov chain Monte Carlo methods represent a powerful approach for the minimization of energy functionals, and more generally for drawing representative samples from probability distributions. The method relies on the generation of a sequence of states. In the Metropolis Hastings algorithm, the next state is accepted or rejected as a simple function of the relative energies of the current state and of that of the next state.

Markov chain Monte Carlo methods have been applied for the problem of segmenting images in guise of so-called Markov random fields, whereby energies are associated with intensity differences between neighbouring pixels, leading to the formation and evolution of boundaries that delimit objects. Because the elementary unit of processing is the pixel, the method tends to be slow.

What we suggest in this proposal is to apply similar methods, not at the pixel level but at the level of entire regions that have been generated using coarse segmentation tools, such as the watershed transform. Indeed, one of the major problems of the watershed transform is over-segmentation that leads to several regions for each object, rather than a single one.

Our methodology can be seen as a model based approach. A rapid survey of the literature did not reveal the availability of a software tool as we envision it. The main goal of the project will be to design a generic software - one that can be tailored on the fly to the particular segmentation problem using a graphical user interface.

In order to demonstrate the generality of the approach, we will aim to progress two separate applications of great biological interest in themselves:

1) Tracking of bacteria in biofilms.
Pseudomonas aeruginosa is an important opportunistic pathogen causing serious and often life-threatening infections in immunocompromised humans. The biological aim is to elucidate the molecular interactions and signal transduction cascades of the complex regulatory system underlying bacterial motility.

In order to enable this research, accurate and reliable segmentation of individual bacteria in time-lapse image data is required. Bacteria have well conserved morphological features that can be exploited to inform and facilitate the image analysis. For example, departure of segmented regions from the average size of bacteria will be associated with an energy term in our framework. Thousands of closely packed bacteria may be present in a single image so fast methods are indeed needed.

2) The segmentation of individual cells in fluorescently stained neurospheres

Neuronal stem cells are seen a very promising innovation to cure neurodegenerative diseases. Under appropriate conditions, neuronal stem cells form self-organised clumps that present a complex organisation in 3 dimensions. Fluorescence staining of the cell nuclei and cell membrane allow distinguishing and counting individual cells. We want to automate this difficult task using our proposed approach. First attempts have highlighted the limitations of more traditional image analysis methods for this problem.

The fellow will also have the opportunity to work on the following projects, related to the main research project:

1. Integration of new segmentation tools into HCA-Vision (Dadong Wang). It is expected that the technology outlined in the preceding paragraphs may be useful for some of the modules in HCA-Vision. Alternately, the technology itself may be turned into an independent module for segmentation problems, much like the product offered by the company Definiens.
2. Porting of the tools to the new GPU Cluster (Luke Domanski). The computational demand on the combinatorial segmentation approach that we are proposing is likely to be high – ideally suited for GPU architectures (many Markov chains can be launched independently and pooled). The fellow will gain valuable experience in this area of growing importance for CSIRO.
3. Develop new analysis techniques for TIRF imaging, particularly in relation to the spatio-temporal statistics of fusion events. (Katarina Mele).

Skills required

A mathematical inclination and an interest in biological imaging.

Skills and knowledge the Fellow will acquire

• Data analysis and mathematical modelling skills
• Skills in high performance computing.

The graduate student will participate in the development of algorithms and methods for operational systems for fine-scale monitoring of urban environments, as part of the Urban Monitor project. The project will be based on the new opportunities presented by digital aerial photography and on the experience of the project partners in development and delivery of satellite-based monitoring systems. Urban and coastal areas are the habitat of choice for most Australians.

They are dynamic, with multiple environmental issues in planning, service provision and resource management and allocation. The advent of high quality digital photography provides for traditional uses as well as “remote sensing” uses such as the monitoring of environmental indicators. A well devised urban monitoring system, based on consistent data and methods will be able to track and communicate changes in features of interest in a way that has not been possible.

Project aims

During the two year period, the successful applicant will work on two distinct problem areas.

1. In the first year, the successful applicant will work on the problems of Geometric constraints and graphical methods for 3D Surface Reconstruction from Images jointly supervised by Peter Caccetta and Xiaoliang Wu. This problem is aimed at improving the quality of surface reconstructions, for example by trying to constrain roof and road reconstructions to be relatively consistent, and properly handling break points due to built features.
2. In the second year, the successful applicant will work on Statistical modeling and analysis of indicators derived from the data in conjunction with ground observations and other spatial data for monitoring the trends in assets such as urban foreshores, wetlands, nature reserves and "bush forever" sites. This work will be jointly supervised by Adrian Baddeley and Peter Caccetta. This phase of the program would introduce the student to spatial statistics while working on a problem of community interest.

Skills required

A degree in a numerically based discipline such as mathematics, statistics, engineering or physics would be an advantage, along with good computing skills.

Skills and knowledge the Fellow will acquire

• Year 1: The applicant will acquire knowledge of advanced image based algorithms and methods, while tackling a problem of significant interest.
• Year 2: The applicant will acquire experience and knowledge of statistical modelling for environmental assessment and research of the use of spatial and temporal indicators.