Current Students

Master of Data Science and Honours in CSSE Research Projects

 

CITS5011 - Project Selection

If you intend to commence your Master of Data Science Research Project in Semester 1, 2019, you need to satisfy the following three criteria:

1. Enrolment in the Master of Data Science (62530)

2.  24 points of Level 4/Level 5 units completed within the course

3. Equivalent of a UWA weighted average mark (WAM) of at least 70 percent 

If you meet the above criteria, please consult the list of available projects below and send your preferences to 'admin-csse@uwa.edu.au' .

CITS4001 - Project Selection

Students enrolled in Honours in Computer Science and Software Engineering undertake a 24 credit points Research Project over two semesters.

 If you are an Honours in CSSE, please consult the list of available projects below and send your preferences to 'admin-csse@uwa.edu.au'. 

 

Prospective students may use the list of available projects below to identify the research areas that our academics are interested in and discuss these or other possible projects with them.

NOTE: Some of the projects require prerequisite skills, such as the completion of a particular unit, as detailed in project description. It is your responsibility to understand and meet the prerequisite skills of any project which you select. If you are unsure, please contact the project supervisor(s).

---------------------------------------------------------------------------------------------------------------------------------

For the list of Projects on offer in Semester 2, 2019 please refer to:

 http://teaching.csse.uwa.edu.au/units/research/

----------------------------------------------------------------------------------------------------------- 

Below are the list of projects for Semester 1, 2019 which might still be available to students in Semester 2, 2019. if interested, please talk to the supervisors:

 

Cardell-Oliver, Rachel, A/Prof

Energy Neutral Sensor Networks

IoT for context-aware decisions

Discipline:  MDS

Prerequisite skills: Essential: Programming (eg R or Python), knowledge of data mining algorithms.  Desirable (one or more of): Computer networks, databases, Web programming, visualization, IoT technologies

Description

Context-awareness can improve predictions based on sensor data.  Internet of things (IoT) technologies can be used to gather context data.  This project involves choosing a sensor dataset that can be enriched by context; testing or adapting algorithms for context discovery, context recognition and context-aware prediction; and developing IoT methods to ingest sensor data for context enrichment

 

Cardell-Oliver, Rachel, A/Prof

Understanding public transport travel patterns

Discipline:  MDS

Prerequisite skills: Essential: Programming (eg R or Python or Java), knowledge of data mining algorithms.  Desirable (one or more of): Databases, Web programming and visualization (e.g. Javascript, D3, visualization libraries), GIS programming.

Description

This project is part of a group supported by the PATREC transport research centre.  Students will build the results of previous projects to understand how Perth’s network is used.  The project involves developing and testing data mining algorithms and visualization of the results.  Summer research scholarships will be offered for this project subject to good performance during the first project semester. Recommended Reading: http://ceed.wa.edu.au/wp-content/uploads/2017/09/17-007-Povey.PATREC.Identifying-Activity-Hubs.pdf

 

Bennamoun, Mohammed, Prof

Boussaid, Farid, Prof

Gavett, Brandon, Prof

Automating neuropsychological  Assessments

Discipline:  MDS

Prerequisite skills: Programming, Machine learning, Computer vision

Description

Neuropsychological assessments are used to detect changes in cognitive functions resulting from brain damage, brain disease, or severe mental illness. The current approach to scoring these neuropsychological tests is fraught with subjectivity and unreliability. In this project, you will investigate advanced  computer vision and machine learning techniques to automate and standardize the process of scoring neuropsychological  tests.

 

Huynh, Du, Dr

Multiple pedestrian tracking in video

Discipline:  MDS

Prerequisite skills: Computer vision, machine learning, Python programming

Description

Multiple pedestrian tracking is an important computer vision task and has many practical applications such as video surveillance, crowd control, pedestrian trajectory prediction, and people counting. A robust multiple pedestrian tracking algorithm should be able to deal with appearance changes of the pedestrians, new pedestrians walking into, and old pedestrians walking out of, the field of view of the camera. The project will require a tracking algorithm to be developed. A suitable pedestrian detector found in the literature can be used with the project.

 

Huynh, Du, Dr

Human action recognition from Kinect video data

Discipline:  MDS

Prerequisite skills: Machine learning, computer vision, Python/Matlab programming

Description

Recognizing human actions from videos is a challenging problem. In this project, the task is to recognize human actions from videos captured by the Kinect sensor using machine learning techniques. The Kinect sensor provides both RGB and depth videos. Some of the Kinect benchmark datasets for human action recognition also have the 3D joint position information of the human subject available. The aim of the project is to investigate using a deep network to learn the features that would allow different actions to be classified and to compare the performance against algorithms that use hand-crafted features, such as HOG (Histogram of Oriented Gradients).

 

Milne George Prof

Simulation model environment for dengue and malaria

Discipline:  MDS

Prerequisite skills: C++ experience

Description

To develop simulation software which will permit a current dengue virus transmission model backbone to be reused as the basis of a new malaria model applicable to our Asia-Pacific region. Specifically, we will develop a malaria model for Honiara in the Solomon Islands. This is of significance as malaria cases are increasing there

 

Mian, Ajmal Prof

Co-supervisors: Akhtar, Naveed, Dr

Adversarial tattoos against face recognition

Discipline:  MDS

Prerequisite skills: The student must have strong programming skills in Python and must be familiar with the Tensor Flow library. He/she should also have reasonable computer system knowledge to run third party code. Knowledge of machine learning will be of additional help.

Description

Deep learning has achieved ground breaking performance on many image classification tasks. However, the learned models are vulnerable to adversarial attacks where subtle changes in the input can lead them to a completely wrong decision. This project will study the possibility of learning small size tattoos that when put on facial images will fool face recognition algorithms to inaccurately identify the person’s identity. This project is mostly of theoretical significance in understanding the vulnerabilities of deep learning. However, it also has applications in privacy. With the growing use of facial recognition systems in commercial use, there is a growing concern about public privacy.

 

Small, Michael, Prof

State evaluation, bifurcation and tipping point transition in time series data

Discipline:  MDS

Prerequisite skills: Mathematics and Statistics, experience with python for numerical computation. Preferably MATH3021 or equivalent.

Description

Time series data is routinely collected across a huge variety of disciplines - from vibration monitoring on the big yellow trucks in the Kimberley to electrocardiographic and electroencephalographic arrhythmia. The aim of this project is to take such data and apply methods from dynamical systems theory and applied mathematics as a form of diagnostic for system health. This may include prediction of tipping point transitions (see Nature Reviews Drug Discovery 16 (2017), 262-272), bifurcation and chaos (Phil. Trans. R. Soc. A 375 (2017), 20160292), or just measuring variance in power spectra or noise. But hopefully something more interesting than the third. Data will come from one of the domains suggested above.

 

Coward, David, A/Prof

An intelligent feeder for a Gamma-Ray Burst database

Discipline:  MDS

Description

Gamma-Ray Bursts (GRBs) are fantastic explosions occurring in the distant universe. Their exact nature starts to be more understood after 50 years of studies: they are the signature of the formation of large black holes through various evolution channels (see Reference 1). Because of their nature, they are the best possible site to study extreme physics that cannot be reproduced in the laboratory.

In its infancy, GRB research was based on an event-by-event method, where each burst was studied in detail and all the properties of the event were drawn out. This led to great successes but also to puzzling enigma. With the launch of the Swift satellite in 2004, this era was closed for an approach based on samples. However, with now more than 25 000 different observations of GRBs publicly available (and about the same amount private), we are facing the usual problems of data mining: there is too much data available to be efficient without specific tools.

To face this problem, a French team forged GRBase, a database of GRBs based on the works previously done by Dr. Gendre (see Reference 2), and some dedicated analysis tools. A first version of this database is available for demonstrations but lacks the completeness: only a subset of all observations was inserted into the database.

Each observation is usually advertised either by a short text message listing the raw results, or by a more complete publication listing the refined data and the properties of the object. In both cases, the information is available in a human readable format, which is not suited for a direct inclusion into an SQL database.

The activity proposed for this Honor is to create a bot which would intercept all the communications of observations and results through a dedicated channel (a web server) and process the content of the message to extract all relevant information and insert it into the database. The student would use advanced AI methods to create this bot, using only GNU components and freeware. The coding would be a modern compiled language like C/C++.

The work will be done at UWA under the supervision of Dr. Gendre, in collaboration with the French team in charge of the servers.

References:

1:https://indico.in2p3.fr/event/13872/contributions/15618/attachments/13043/15984/DTURPIN_SVOM2017.pdf

 

Liu, Wei, Dr

Co-supervisors: Holden, Eun-Jung, Prof

Knowledge Graph Embeddings for Geological Survey Analysis

Disciplines:  MDS, CSSE Honours

Prerequisite skills: Ideally should have done CITS5008 Machine Learning or equivalent.

Description

Knowledge Graphs (KG) and their embeddings have been proven to capture the semantic information of concepts and allow for in  inferences such as link prediction, relation extraction as well as question answering. In this research, we will build a text-enhanced knowledge graph embedding model for a geological survey report database, in a similar fashion to TEKE [2], and compare that with other knowledge graph embedding models that are capable of incorporating textual descriptions, e.g. description-embodied knowledge representation learning (DKRL) [1].

Reference:

[1] R. Xie, Z. Liu, J. Jia, H. Luan, M. Sun, "Representation learning of knowledge graphs with entity descriptions", Proc. 30th AAAI Conf. Artif. Intell., pp. 2659-2665, 2016.

[2] Z. Wang, J. Li, "Text-enhanced representation learning for knowledge graph", Proc. 25th Int. Joint Conf. Artif. Intell., pp. 1293-1299, 2016

 

Liu, Wei, Dr

Using computational data mining to transform information into a manageable knowledge base for researchers

Discipline:  MDS, CSSE Honours

Description

Background: The ever-growing body of online information has highlighted new challenges in extracting relevant meaningful content. Across disciplines the discovery and summary of published research is a daily task for scientists and commercial researchers in clinical, biological and technological research. When done systematically and shared, such knowledge bases have shown extremely useful in biology, as demonstrated by SUBA (http://suba.live/) and cropPAL, 2 (http://crop-pal.org/). Evidence shows that current practice is too time-intensive and expensive. Better software would save labour cost and increase productivity as well as the enjoyment of learning. 

A number of search and data mining strategies exist that show promise for integration into a content and career-strategic algorithm that generates a research knowledge base for individual research staff and students.

 

Outline: This applied project will make use of the unique data resources available from the Centre of Plant Energy Biology at UWA and focus on experimenting with different strategies around data mining, machine learning and literature-social network modelling algorithms for the purpose of developing an optimized way to build a knowledge database for a given biological study subject.

Joining the team gives the student a unique opportunity to work with computer scientists, plant biologists, data scientists, librarians and computational biologists. The generated output will be immediately applicable to local scientists at UWA.

Methodology:

The methodology includes but is not limited to

•          Automated collection of text and data from the internet for analysis

•          Constructing association networks using collected data from internet resources and from local database repositories

•          Establishing selection criteria for document classification and clustering

•          Statistical testing of findings

•          Language processing and content analysis

•          Interdisciplinary team/ networking strategies for taking research outcome to direct users

 

Hassan, Ghulam Mubashar, Dr

Co-supervisors: Datta, Amitava, Prof

Detection of health related issues using social media

Discipline:  MDS

Prerequisite skills: Strong programming skills: CITS2401 with HD or CITS1401 with HD/D or equivalent

 

Description

Social media is very popular to monitor different aspects of life. It will be interesting to analyse how health related issues can be monitored using social media data. The users of social medial are regularly updating their status which can help to determine health issues related to users or arising health issues.

 

This project will require students to use artificial intelligence and data mining techniques to analyse data and retrieve appropriate information from the data.

 

Wise, Michael, A/ Professor

Co-supervisors: Liu, Wei, Dr

Keyword Extraction and Clustering of Research Ethics Cases

Disciplines:  MDS, CSSE Honours

Description

COPE, the Committee on Publication Ethics http://publicationethics.org/ , is a worldwide organisation that provides advice and best-practice guidelines regarding publication ethics to its 12,000 members, each an academic journal or publisher. Among the services COPE provides, member journals are able to submit details of cases, which are discussed in open Forum of members or by COPE Council (i.e. the governing board). Out of these discussions, advice is distilled and transmitted to the case submitter. Having operated in this manner for nearly 10 years, there is a considerable database of anonymized cases (Researcher A, Prof B, Journal C, etc). A unique set of 563 cases is available as a .csv file, together with some metadata for each case.

Ignoring the metadata for the time being, the aim of this project is to automatically discover keywords around which cases can be clustered. The cases will need to be divided in to the usual training and test subsets. The set of keywords can be compared to a hand-crafted set of keywords created by COPE council members in a project

undertaken some years ago.

 

(The project has been cleared by UWA Human Research Ethics Office as been exempt from review.)

 

Bekki, Kenji, Dr

Co-supervisors: Hassan, Ghulam Mubashar, Dr

Estimating the total masses of dark matter in galaxies with deep learning

Disciplines:  MDS

Prerequisite skills: Python programming, basic knowledge about deep learning

Description

Dark matter is the major component of the Galaxy: more than 90% of the matter is believed to be ``invisible matter’’. In this project, students try to estimate the total amount of this mysterious dark matter within galaxies by applying deep learning to a large amount of data on dark matter from computer simulations of galaxies. More specifically, the simulated spatial distributions and kinematics of star clusters within the outer parts of galaxies will be used to estimate the total masses of dark matter within galaxies through deep learning algorithms: this is a regression problem. First, convolutional neural networks will be constructed to accurately predict the total mass of dark matter in a galaxy.  Then the CNNs will be applied to real observational data to estimate the missing (dark matter) masses of galaxies.

 

Wu, Chen, Dr

Co-supervisors: Wong, Ivy, Dr

Extracting radio galaxy classifications from citizen science forum hashtags

Disciplines:  MDS

Prerequisite skills: Experience with MongoDB is desirable but not a pre-requisite.

Programming skills in data science (e.g. Python, or R) will be good

Description

Radio galaxies are galaxies that host central supermassive black holes which emit radio jets as a result of matter accreting into the black hole. Over time, the jet components become dissociated into multiple disconnect components which can extend over far away from the host galaxy. Radio Galaxy Zoo is an online citizen science project aimed at matching up the associated radio source components and identifying the host galaxy. Hashtags are used in the forum to describe these radio galaxies. The aim of this project is to study the collection of hashtags, derive statistical patterns and further our understanding of radio galaxy evolution.

 

Wu, Chen, Dr

Co-supervisors: Wong, Ivy, Dr

Building knowledge graphs for radio galaxies

Discipline:  MDS

Prerequisite skills: Programming skills in data science (e.g. Python, or R)

Basic knowledge in machine learning

Familiarity with natural language processing is a plus but not essential

Description

Radio Galaxy Zoo is an online citizen science project aimed at characterising and identifying radio galaxies of complex structures through crowd-sourcing. A knowledge graph allows astrophysicists and data processing algorithms to tap into the social interactions on the radio galaxy zoom platform in order to search, classify and reason over key properties of radio galaxies and their associated physical processes. To do this, a knowledge graph can be built and evolved in an semi-automated fashion by using some exciting machine learning techniques such as word embedding, link prediction, etc.

 

Corrêa, Débora, Dr

Co-supervisors: Walker, David, Dr

      Small, Michael, Prof

Concept drift detection in time series using complex networks

Discipline:  MDS

Prerequisite skills: Experience with python, Matlab or R.

Basic knowledge of statistics.

Description

The representation of scalar time series as a complex network has proved to be useful to derive insights about the underlying process generating the data. The majority of existing techniques, however, do not scale well with the length of the time series, leading to very large networks. Our group has developed a strategy in which a compression algorithm is applied to the time series and its output is used to generate a complex network. The data compression algorithm exploits recurrences in the time series leading to more compact representations of the complex network coined a compression network.  In this project, we want to generate compression networks in a windowing basis in order to detect concept drift in time series. The main objective of the project is to develop suitable strategies that are capable of relating changes in compression network topologies to changes in the behaviour of the original time series. The successful application of this framework can be directly applied to many industrial scenarios, for instance, anomaly detection and predictive analysis

 

Corrêa, Débora, Dr

Co-supervisors: Walker, David, Dr

                             Small, Michael, Prof

Comparison of recurrent neural networks for time series prediction

Discipline:  MDS

Prerequisite skills: Experience with python, Matlab or R.

Basic knowledge of statistics.

Basic knowledge of machine learning.

Description

Recurrent neural networks (RNNs) are types of artificial neural networks (ANNs) which include feedback connections, allowing a context to persist across iterations of the network. They have been successfully used for time series prediction and classification, when temporal dynamics connecting data samples are important. From the more general category of RNNs, two paradigms have emerged: the memory cell RNNs, which are trained with the traditional back-propagation algorithm; and the echo-state networks, which offer a more simplified training procedure. This project proposes a comparative investigation of these two paradigms, assessing their performance for time-series related tasks under different scenarios, such as noise, nonstationarity and prediction horizon.

 

Portes dos Santos, Leonardo Luiz, Dr

Co-supervisors: Correa, Debora, Dr

                             Small, Michael, Prof

Subsurface Big Data Visualisation and Analysis: Oil/Gas Reservoirs

Discipline:  MDS

Prerequisite skills:

- Experience with Python or R (Jupyter notebook is a plus).

- Able to explain concepts in different levels of complexity (e.g.,

teaching experience, tutoring etc).

Description

Decision making, in different levels in the oil/gas industry, depends on appropriate insights gathered from subsurface data and modeling. Advancements in computing power and subsurface technology led to a large amount of data (both in size and type), for which this industry standard methods/approaches are not enough to cope with. Hence, the understanding of subsurface professionals could not fully benefit from the data they already have, and new approaches for visualization and analysis are urgently welcome.

In this aspect, one high advanced field is the “big data” approach for gene expression, for which our group has contributed through a complex network analysis paradigm in the context of cancer immunotherapy. Our proposal is the application of this know-how from genetics into typical contexts of the oil/gas industry: (i) well placement optimization and (ii) geological uncertainty quantification. The main objectives are the development of data visualization/analysis methods and tools for both (a) summarizing results and (b) assessment of reservoir input and output correlation. The directive is a practical approach, to actual problems of this industry, with potential to be used by subsurface professionals without any training in nonlinear dynamics, network science or big data.

The selected students should cooperate as a team on the conceptual level, but each will choose and develop his own investigation as a combination of items (i, ii) and (a,b): for example, one student may decide to investigate how to summarize results for well placement optimization. Besides that, it is expected, and it will be fully supported by their Supervisors, the development of partnerships with oil/gas industries in Perth (which already show high interest for this topics).

 

Zaitouny, Ayham, Dr

Co-supervisors: Correa, Debora, Dr

                             Correa, Debora, Dr

                              Small, Michael, Prof

Data mining for gene expression analysis to reveal cancer dynamics

Discipline:  MDS

Prerequisite skills: Coding- Data mining

Description

The project is based on collaboration with UWA-Harry Perkins institute and Telethon Kids institute, the project is to develop an immunotherapy for cancer by using genes RNA sequencing. Microarrays and RNA sequencing technologies are game-changers in biological science. However, the data provided is highly complicated and requires sophisticated algorithms to be analysed. We are focusing on developing mathematical and dynamical algorithms to analyse the genes RNA sequences data, namely complex network, machine learning and tipping points detection algorithms in dynamical diseases. Working with clinicians we aim to contribute to the development of new treatment regimes for dynamical diseases. This should provide significant benefit to many biological areas specifically those use genes RNA sequences. In this project, the nominated student will help in running the data analysis and visualisation.

 

Reynolds, Mark, A/Prof

Co-supervisors: Huynh, Du, Dr

Industry Partner:  Dr Richard Thomas (consultant), Angela Saunders (Bush Heritage)

Bat Call Identification via Machine Learning

Discipline:  MDS

Prerequisite skills: Python (good), some knowledge of machine learning

Description

Bats are useful indicator species in ecological surveys. Typically a device will record ultrasonic echolocation calls in the field and the subsequent data will be analysed to identify the bat species present. This is a laborious process that is amenable to machine learning. One such proprietary system has been used successfully to classify several years of calls in the South Coast region of WA.

However some bat species, especially of the genus nyctophilus, are not amenable to the zero crossing techniques commonly used. McKenzie and Bullen (2003, 2009, 2012) have shown that the sharpness quotient, Q, of the fundamental harmonic and the characteristic frequency of the bat call cluster rather distinctly between different species of bats including nyctophilus.

The aim of this project is to examine whether similar techniques might be used for machine learning of call identification for the bats of the South Coast region.

You would be provided with full spectrum recordings covering several years in WAC/WAV files plus zero crossing analysis data and probable bat identification.

There would be a requirement to complete a Bush Heritage Australia research project form which details IP and the like.

 

Reynolds, Mark, A/Prof

Co-supervisors: Huynh, Du, Dr

                             Liu, Wei, Dr

Industry Partner:  Surgical Realities Pty Ltd

Development of an intelligent surgical counting system for improved patient safety

Discipline:  MDS

Prerequisite skills: Python (good), some knowledge of image processing

Description

Ensuring patients remain free of unintended retained foreign bodies is a primary responsibility of perioperative nurses and surgical technologists.  AORN's "Recommended practices for prevention of retained surgical items" provides a standard protocol for the prevention of retained foreign bodies: however, these incidents continue to occur despite hospital polices and AORN recommended practice guidelines for their prevention. The surgical count plays a vital role in preventing retained items and protecting patients from harm; however, incorrect surgical counts in the operating room (OR) are common and factors contributing to their occurrence remain poorly understood. This project will investigate the literature, define the contributing error factors and specifically explore the use of computer vision, machine learning for object tracking, recognition and counting to augment the current manual surgical count in accordance with the recommended best practice guidelines.  It will involve familiarisation with the OR environment and equipment and require working closely with the OR surgical team and perioperative nurses.

 

Reynolds, Mark, A/Prof

Co-supervisors: Male, Sally, Prof

       Liu, Wei, Dr

Industry Partner:  UWA Faculty of Engineering and Mathematical Sciences

Machine learning to identify "non-constructive" student feedback

Discipline:  MDS

Prerequisite skills: Python (good), some knowledge of machine learning

Description

Teachers, lecturers and presenters increasingly use formal techniques to ask for feedback and evaluation on their activities. This is valuable as it helps to improve the work and make it more engaging and useful to the audience. Some of the most valuable feedback, comments and suggestions come in the form of free text where that facility is offered to the audience. Unfortunately, especially in the case of anonymous feedback, free text fields can misused to contain non-constructive, inappropriate, irrelevant or abusive content. In order to protect the presenter, or other readers, and to allow more efficient collection and analysis of useful feedback it would be helpful if such comments could be automatically identified, or if at least suspicious sections could be flagged.

Machine learning and other state of the art text processing algorithms can be investigated as potential parts of a solution to this problem. The aim would be to see what can be produced in the way of a simple initial prototype system.

 

Reynolds, Mark, A/Prof

Causal Reasoning

DisciplinesMDS, CSSE Hons

Prerequisite skills: Python (good), some knowledge of machine learning

Description

Suppose that we are given two parallel time series, X and Y. Suppose that we want to predict Y from X. There are subtle differences in this task depending on whether we think that X causes Y or that Y causes X, or neither. This project is to develop general machine learning models that can work well at the task across these cases.

 

Reynolds, Mark, A/Prof

Co-supervisors: Barbour, Liz, Dr (CRC Honey Bee Products)

      Huynh, Du, Dr

Industry Partner:  CRC Honey Bee Products

Bee Identification and Tracking in Video

Disciplines:  MDS, CSSE Honours

Prerequisite skills: Programming (familiarity with Python)

Description

Understanding bee behaviour is important for ecological and economic reasons. In the Australian Government funded Cooperative Research Centre (CRC) for Honey Bee Products, researchers record videos of bee activities near flowers in the Australian bush. Currently useful information such as bee species identification, bee numbers and bee movement between flowers is extracted from the recording by human observers. This project will use current UWA CSSE video processing tracking techniques and machine learning identification algorithms to attempt to automate most of the information extraction. Related work will explore the geographical spatial distribution of bee activities in the areas under study. The student will work closely with CRC scientists.

 

Reynolds, Mark, A/Prof

Co-supervisors: Hu, Yuxia, Prof

      Hassan, Ghulum Mubashar, Dr

      While, Lyndon, Dr

Industry Partner:  Main Roads WA and other state government transport agencies

Optimisation of scheduling of maintenance using genetic algorithm approaches

DisciplinesMDS, CSSE Honours

Prerequisite skills: Good Programming (familiarity with Python)

Description

Main Roads WA is interested with working with UWA on a range of computational modelling, visualisation, machine learning, AI and Data Science projects to do with traffic modelling, traffic monitoring and road maintenance.

 This topic: Optimisation of scheduling of maintenance using genetic algorithm approaches.

 The exact projects will be developed in consultation with Main Roads WA, the students and the supervisors.

 

Reynolds, Mark, A/Prof

Co-supervisors: Huynh, Du, Dr

Industry Partner:  Main Roads WA

Detecting road deterioration from dashcam video footage

Disciplines:  MDS, CSSE Honours

Prerequisite skills: Good Programming (familiarity with Python)

Description

Main Roads WA is interested with working with UWA on a range of computational modelling, visualisation, machine learning, AI and Data Science projects to do with traffic modelling, traffic monitoring and road maintenance.

 Some of the topics to be tackled include:

 This topic: Detecting road deterioration from dashcam video footage (machine learning).

 The exact projects will be developed in consultation with Main Roads WA, the students and the supervisors.

 

Reynolds, Mark, A/Prof

Co-supervisors: Huynh, Du, Dr

       Hassan, Ghulum Mubashar, Dr

Industry Partner:  Main Roads WA

Working with machine learning on very unbalanced image data sets

Disciplines:  MDS, CSSE Honours

Prerequisite skills: Good Programming (familiarity with Python)

Description

Main Roads WA is interested with working with UWA on a range of computational modelling, visualisation, machine learning, AI and Data Science projects to do with traffic modelling, traffic monitoring and road maintenance.

 Some of the topics to be tackled include:

 This topic:

Working with machine learning on very unbalanced image data sets. These are images of roads in various conditions.

 The exact projects will be developed in consultation with Main Roads WA, the students and the supervisors.

 

Reynolds, Mark, A/Prof

Co-supervisors: Huynh, Du, Dr

       Sun, Chao, Dr

Industry Partner:  Main Roads WA

Detection of patterns of vehicle behaviour from traffic camera video

Disciplines:  MDS, CSSE Honours

Prerequisite skills: Good Programming (familiarity with Python)

Description

Main Roads WA is interested with working with UWA on a range of computational modelling, visualisation, machine learning, AI and Data Science projects to do with traffic modelling, traffic monitoring and road maintenance.

 This topic:

Detection of patterns of vehicle behaviour from traffic camera video

 The exact projects will be developed in consultation with Main Roads WA, the students and the supervisors.

 

Reynolds, Mark, A/Prof

Co-supervisors: Huynh, Du, Dr

       Sun, Chao, Dr

Industry Partner:  Main Roads WA

Predicting road congestion ahead of time

Disciplines:  MDS, CSSE Honours

Prerequisite skills: Good Programming (familiarity with Python)

Description

Main Roads WA is interested with working with UWA on a range of computational modelling, visualisation, machine learning, AI and Data Science projects to do with traffic modelling, traffic monitoring and road maintenance.

 This topic:

Detection of patterns of vehicle behaviour from traffic camera video

 The exact projects will be developed in consultation with Main Roads WA, the students and the supervisors.

 

Reynolds, Mark, A/Prof

Co-supervisors: Sun, Chao, Dr

      McDonald, Chris, Dr

Machine learning recognition of transitions in a multi-leg journey using sound data and movement sensors

Disciplines:  MDS, CSSE Honours

Prerequisite skills: Good Programming (familiarity with Python or Java, willing to work in C#)

Description

A start-up company developed an app called Freewheeler, which rewards people for making the right (healthy, non-polluting) travel choice such as cycling and walking.

 They have a prototype working but want more improvements.

 Potential research topics involve recognizing vehicle types by sound (machine learning with sound data), recognizing travel mode via mobile phone movement sensors, machine learning recognition of transitions in a multi-leg journey.

 The exact projects will be developed in consultation with the company, the students and the supervisors. An IP agreement will need to be signed.

 

Reynolds, Mark, A/Prof

Co-supervisors: Hong, Jin, Dr

Industry Partner: WA State Government Department of Treasury

Application of AI and Machine Learning in Cybersecurity

Disciplines:  MDS, CSSE Honours

Prerequisite skills: Good Programming (familiarity with Python)

Description

Cybersecurity is an escalating challenge.  In the energy sector, State and Federal agencies, market bodies and energy companies are all stepping up their cybersecurity actions.  Policy development and coordination is also being stepped up.  This project, in collaboration with the WA Treasury, will investigate applications of Artificial Intelligence and Machine Learning which might show how they can play a significant role in this in future

 

Reynolds, Mark, A/Prof

Industry Partner: WA State Government Department of Treasury

Heat maps of demand for primary school locations

Disciplines:  MDS, CSSE Honours

Prerequisite skills:

Good Programming (familiarity with Python)

Description

Treasury is interested with working with us on a range of computational modelling, visualisation, machine learning, AI and Data Science projects.

Some of the topics to be tackled include:

Schools

Projections and heat maps of demand for primary school locations to assist the Department of Education to prioritise its asset investment program.

 

Reynolds, Mark, A/Prof

Industry Partner: WA State Government Department of Treasury

Modelling customer demand for electricity

Disciplines:  MDS, CSSE Honours

Prerequisite skills:

Good Programming (familiarity with Python)

Description

The Public Utilities Office (PUO) in Treasury contracts a lot of sophisticated modelling to look ahead to find the most cost effective way to meet customer demand for electricity generation and transmission.  Supply must match demand exactly, over the whole network, from moment to moment, without interruption.  A project could examine how AI could improve various elements of modelling, improving accuracy (as it has done for commercial supply chain forecasting) and potentially save electricity bills

 

Reynolds, Mark, A/Prof

Industry Partner: WA State Government Department of Treasury

Real time management of electricity generation systems

Disciplines:  MDS, CSSE Honours

Prerequisite skills:

Good Programming (familiarity with Python)

Description

Treasury is interested with working with us on a range of computational modelling, visualisation, machine learning, AI and Data Science projects.

Electricity generation plant and electricity networks are becoming increasing automated and far more complex.  Within a few years, AI systems could play a major role in the real time management of these systems.  The State owned electricity network business, Western Power, is already engaged with and planning for AI.  A project could be developed in partnership with Western Power or another of the GTEs.

The exact projects will be developed in consultation with the Treasury, the students and the supervisors.

 

Reynolds, Mark, A/Prof

Co-supervisors: Coward, D., Prof Howell, E., Prof and Gendre, B. , Prof (Physics)

An AI bot for a satellite database

Disciplines:  MDS, CSSE Honours

Prerequisite skills: Good Programming (familiarity with Python)

Description

The activity proposed for this project is to create a bot which would intercept all the communications of observations and results through a dedicated channel (a web server) and process the content of the message to extract all relevant information and insert it into the database. The student would use advanced AI methods to create this bot, using only GNU components and freeware. The coding would be a modern compiled language like C/C++ or python.

The work will be done at UWA under the supervision of Dr. Gendre, in collaboration with the French team in charge of the servers.

 

Reynolds, Mark, A/Prof

Co-supervisors: Coward, D., Prof Howell, E., Prof and Gendre, B. , Prof (Physics)

Smart scheduling for a robotic telescope

Disciplines:  MDS, CSSE Honours

Prerequisite skills: Good Programming (familiarity with Python)

Description

The project will employ data from gravitational wave detectors to test algorithms for targeting robotic telescope searches for real sources. A smart program will be developed to optimise the pointing and scheduling of telescopes using probability sky maps.

 

Reynolds, Mark, A/Prof

Co-supervisors: Coward, D., Prof and Howell, E., Prof (Physics)

A machine learning data mining search for correlations in high energy satellite data

Disciplines:  MDS, CSSE Honours

Prerequisite skills: Good Programming (familiarity with Python)

Description

The project will employ a large data base to search for correlations between parameters observed by NASA satellites FERMI & BATSE. The correlations will be used to re-interpret and understand the astrophysics powering cosmic explosions.

 

Reynolds, Mark, A/Prof

Co-supervisors: Bucks, Romola S, Prof (Psychology)

      Mike Weinborn, Maria Pushpanathan

A machine learning algorithm for assisting diagnosis of mild cognitive impairment

Disciplines:  MDS, CSSE Honours

Prerequisite skills: Good Programming (familiarity with Python)

Description

The Healthy Ageing Research Program (HARP) is a longitudinal study of community adults aged 50+ exploring cognition, psychological well-being and quality of life, and their predictors.

 We need an algorithm to apply diagnostic criteria for mild cognitive impairment or dementia to cognitive data, looking up the values in a database, then applying the algorithm (with expert input from us to amend it) to produce a diagnosis per participant, per time point assessed. This will need to be able to handle missing data.

 

Reynolds, Mark, A/Prof

Co-supervisors: Bucks, Romola S, Prof (Psychology)

                 Mike Weinborn, Maria Pushpanathan

Textual analysis of free fields containing treatment data

Disciplines:  MDS, CSSE Honours

Prerequisite skills: Good Programming (familiarity with Python)

Description

The Healthy Ageing Research Program (HARP) is a longitudinal study of community adults aged 50+ exploring cognition, psychological well-being and quality of life, and their predictors.

 The project will involve textual analysis of free fields containing data on a) medications, b) medical conditions, to categorise them by type and count the numbers of each for later analysis.

 

Reynolds, Mark, A/Prof

Co-supervisors: Sun, Chao, Dr

A data-driven approach to designing demand responsive public transport

Disciplines:  MDS, CSSE Honours

Prerequisite skills: Python or R for data science,  Machine Learning Optimisation

Description

Traditional Public Transport (PT) services (e.g. bus routes and schedules) have been determined by transit agencies using a top-down approach that is mostly based on past experiences and rough estimations. This often results in low level of service and leads to low patronage and loss of revenue. In recent years, demand responsive PT has been put forward as a potential solution. As it is still in its infancy, it is still unclear how such services should be best organised or designed as there are many different ways to do it. This project will investigate how to leverage multiple large datasets (including Perth's superb SmartRider data) to optimise demand responsive PT service provision for Perth.

 

Reynolds, Mark, A/Prof

Co-supervisors: Sun, Chao, Dr

Industry Partner: PATREC

Optimising SmartRider data visualisation

Disciplines:  MDS, CSSE Honours

Prerequisite skills: Python or R for data science, Machine Learning Optimisation

Description

Perth's SmartRider is a large data source that contains rich information about people's travel behaviour. Visualising such spatial-temporal data will help discover patterns that are otherwise not obvious. It also serves as an effective communication tool to draw public interest in engaging discussion around public transport. This will be a continuation of a commercial project that the Planning and Transport Research Centre (PATREC) is currently finishing. In such projects, there are many design variables that need to be determined by the team and there is no guarantee that they are the optimal combination that will achieve the best outcome. The idea is to use principles of genetic algorithm to evolve the best parameter set by using test audience as the fitness function. It is also possible to evolve parameterised music based on the data to enhance communication.

 

 

Reynolds, Mark, A/Prof

Co-supervisors: Hassan, Ghulam, Dr

Exam Marking of Diagrams via Machine Learning

Discipline:  MDS, Honours in CSSE, Maths, EE

Prerequisite skills: Python (good), some knowledge of machine learning

Description

Marking of exams is one of the tasks of lecturers and teachers that would be most useful to automate. There already exist systems that work well in managing online multi-choice quizzes or testing students programming expertise. There are also a few prototypes of systems using machine learning (ML) designed to read and grade hand-written essays and short text answers. In many disciplines, though, diagrams and formulas play an important part in exam answers.

This project aims to produce a prototype hand-written diagram marking system. The idea is to use ML to recognize parts of a diagram and reconstruct the diagram in a formal notation. The formal version of the diagram can then be graded using a rule-based system or another ML system.

Actual graded student exam answers using diagrams will be available.

 

 

Reynolds, Mark, A/Prof

Co-supervisors: Wang, Jingbo, Prof

Logic via Quantum Computing

Discipline:  Honours in CSSE, Physics, Maths, EE

Prerequisite skills: Good linear algebra skills

Description

Can quantum computers calculate anything faster than classical computers? A famous result from 1994 shows that theoretically they can factor integers exponentially faster than any known classical algorithm. But that does not prove that classical computers are slower: there might be classical methods as yet unknown which solve this problem.

A new 2018 result from an IBM research lab finds a class of problems and shows that a certain type of quantum algorithm, fixed circuit depth ones, can solve such problems. However, no fixed circuit depth classical algorithm can solve the problems.

See the blog and video at https://www.ibm.com/blogs/research/2018/10/quantum-advantage-2/

One important fixed circuit depth problem is 3-SAT which is a famous NP-complete decision problem. This is the problem of determining whether a Boolean, or classical propositional  logic formula (in a certain restricted format) is satisfiable, or could be made true by choice of truth values of its propositional atoms.

This project aims to see if any speed-up can be hoped for in using Quantum Computing on related propositional logic search algorithms.


Wise, Michael, A/Prof

Detection of Potential Ghost-writing Through Stylistic Analysis

Discipline:  MDS

Description

UWA, like every university around the country (probably around the planet) is very worried about ghost-written submissions for assignment.

It's also known as contract cheating.  Whatever you call it, it's about getting someone else to do your work, but submitting it as if it was only your work. In this case I'm only concerned here with essays. The incidence is believed to be low, but it's clearly not a good thing.

What I would like you to do is write a program in Python that uses stylistic analysis to compare essay texts with a corpus of other writing allegedly from the same person, not on the basis of plagiarism (where expect texts to be different), but based on metrics related to writing style (where you expect the styles to be largely the same). In this case, the test texts are essays that CITS3200 students submitted, and the comparison texts are concatenations  of 3 structured reflections that they have also supplied.

To test the methods, for a small number of cases, the Reflections of on person will be associated (randomly) with a different person.

Alternatively, the Reflections can be shuffled and associated with different essays, and the aim will be to find the most likely partner.

Ethics

Please note that the texts are being used here only with the explicit agreement of the students who submitted them.

The texts have been completely anonymised. File names have been changed from student numbers to generic names, and the corresponding reflections concatentations have been changed to generic names. In addition, given that all the names of students in the class was known, all the texts will have the names of people in the class replaced with generic names(including  first names if they appeared separately) and, of course, surnames. An application has been made to the UWA Human Research Ethics office for ethics clearance for this project


Foong, Rachel, Dr

Co-supervisors: Rauschert, Sebastian, Dr

   Broadhurst, David, Prof

   Stevenson , Paul, Mr

Start Date: July 2019 

Research Focus Area Chronic & Severe Diseases

Research Group: Children’s Lung Health

Personalised, machine learning based prediction of asthma and allergies in Western Australia

Discipline: PhD, MDS and Honours in CSSE

 Prerequisite skills:

·         Have an Honours degree or equivalent in statistics or other data analysis qualifications

·         Eligible to enrol in a PhD at a University

·         Self-management and high personal motivation

·         Excellent communication and interpersonal skills

·         Knowledge of a programming language (preferably R, Python or related data analysis and machine learning tools)

Some experience with the UNIX computational environment and/or cloud computing

Project Outline         

Asthma is the most common chronic lung disease of childhood. In Western Australia, nearly 9% of children report having a doctor diagnosis of asthma. Asthma diagnosis is difficult in young children. Therefore, there has been a worldwide effort to develop ways to identify asthma risk as early as possible in order to prevent disease. This project aims to show that asthma and allergies in individuals can be predicted before it occurs based on individual family history and information on the early environment. The aim is to create personalized prediction scores for the development of asthma and allergy by using machine learning, which will help in better understanding, preventing and managing asthma.

 The Raine Study will be utilised to develop the prediction scores. For this study over 2800 pregnant women were recruited in Perth, WA and asthma and allergies have been well-studied in this cohort at 5/6-years, 13/14-years and 22/23-years in children. In addition to questionnaire data, the Raine study has lung function and immunological measurements from participants which can be used to generate prediction scores. Furthermore, genetic and epigenetic data is available for more than 1000 study participants, which may be assessed in relation to the predictive modelling. Once prediction scores have been developed, the findings will be validated in other birth cohorts within Australia and within international birth cohort studies.

 The findings from this study will identify risk factors in early-life for asthma and allergies. In the era of personalized medicine, the ability to undertake an individualized risk assessment based on familial history and past exposures may present an attractive means for targeted approaches for early intervention and prevention. Furthermore, a predictive model of asthma and allergies based on questionnaire data and family history may help doctors make decisions quickly in a cost-efficient way. It can improve diagnosis of asthma and allergy and can be helpful in disease management.

 Full scholarship offered by project group

 For more information, please contact:

Dr Rachel Foong

08 6319 1626

Rachel.Foong@telethonkids.org.au