Project offerings for Semester 1 2016

Click on the supervisor's name for a description of the projects they are offering.

Projects will be added over the coming weeks.


Supervisor Project Credit points
Weidong (Tom) Cai       Cloud-Based Large-Scale 3D Biomedical Image Processing System 12
4D Neuroimage Computing for Tracking Tissue Changes 12
Feature-Centric Content Analysis in Biomedical Images 12 or 18
Contextual Feature Representation for Image Pattern Classification 12 or 18
Intelligent 3D Single Neuron Reconstruction 18
Neuroimaging Computing for Early Detection of Dementia 18
Context Modeling for Medical Image Retrieval 18
Vera Chung    Video Object tracking for visiting applications 18
Moving object detection for video surveillance 18
Study of behaviour detection algorithm for objects in video streams 18
Deep Neural Networks for Object Detection 18
Joachim Gudmundsson Role-assignment of football players 12 or 18
Dominant regions of football players 12 or 18
Classifying attacks in football 18
Classifying passes in football 18
Bryn Jeffries    In-database motif detection of time series data 18
Feature extraction of speech task data 18
App for actigraphy-assisted sleep diaries (various projects) 12
Kevin Kuan    Mobile Web Beer Game 12
What Makes an Online Consumer Review Helpful? 12 or 18
Emotional Contagion in Online Social Networks 12 or 18
Effect of Information Overload in Online Decision Making 18
David Lowe
   
Augmented reality remotely-accessed labs 12 or 18
Architecture for collaborative remotely-accessed labs 12 or 18
Remote control of physical systems: the role of context 12 or 18
Simulation of chronic pain through control of pressure gloves 12 or 18
Josiah Poon Information extraction from Discharge Summary 12
Bernhard Scholz   Build and Test Harness for Soufflé 12
Language Bindings for Soufflé 12 or 18
Profiler & Debugger for Soufflé 12 or 18
Masa Takatsuka   Remote Interactive 3D graphics 12 or 18
Adaptive Self-Organising Map for streaming data 12 or 18
Multi-threaded 3D game engine 12 or 18
Xiuying Wang, Jianlong Zhou and David Feng     Deep learning approach for medical image segmentation 18
Assessment of quality of medical image segmentation with physiological measurements 18
Investigation of user trust on medical image segmentation 18
Visual analytics of multi-stream life data for health monitoring 18
Content-oriented deformable image registration 18
Zhiyong Wang Multimedia Data Summarization 12 or 18
Predictive Analytics of Big Time Series Data 18
Human Motion Analysis, Modeling, Animation, and Synthesis 12 or 18
Intelligent multi-video editing 12
Food recognition 12
View synthesis for VR/AR 12
Bing Zhou Computing and visualising transcription factor interaction networks 18
Ying Zhou Visualizing Wikipedia Co-authorship Network 12
Albert Zomaya and MohamadReza Hoseiny      Energy-Efficient Qos-Aware Dynamic Workload Consolidation in Virtualized Cloud Datacentre 12 or 18
Building a flexible event-driven framework for cloud based applications 12 or 18
Designing the optimal data model of large-scale applications in Google App Engine 12 or 18
Optimal scheduling of bag-of-task application in hybrid cloud environment in the presence of heavy-tailed tasks with bursty traffic arrival pattern 12 or 18
Running Workflow application (Montage as a case study) in Cloud 12 or 18
PANDA: a PAreto Near-optimal Deterministic Algorithm for Scaling Out Large-Scale Bag-of-Tasks Applications Across Multiple Clouds 12 or 18
Albert Zomaya and Wei Li     Addressing Interoperability in the Internet of Things 12 or 18
Computation Offload in Fog Computing 12 or 18
Handling Big Data-Streams in the Internet of Things 12 or 18
Task Scheduling Algorithms for the Internet of Things 12 or 18
Virtualization of Wireless Sensor Networks 12 or 18

 

Project supervised by Weidong (Tom) Cai

Cloud-Based Large-Scale 3D Biomedical Image Processing System (12cp)
The software platforms for processing biomedical images are key components of health care practices and biomedical research. Most of the existing biomedical frameworks are desktop-based diversifying significantly with their user experience and functionalities. The local biomedical computing software have many disadvantages: (1) the local systems are hardly portable between different hardware platforms; (2) they are hard to be kept up-to-date with newly proposed algorithms due to the difficulty of maintenance; (3) the locally stored biomedical images and the processed results are at a high risk of losing or damage; (4) they require expensive hardware support to achieve a reasonable computing performance, etc. This project aims to build a cloud-based biomedical image computing system that is accessible via the web browsers. This system is expected to be able to visualise, process, analyse and store the 3D biomedical images. To make a solid contribution to the biomedical image computing society, the final framework is expected to be fast, scalable and flexible with different biomedical image processing pipelines.

Preferred Skills:
Knowledge of web programming, cloud platforms, databases, basics of 3D graphics

Preferred Programming Languages: C/C++, Python, javascript

4D Neuroimage Computing for Tracking Tissue Changes (12cp)
Non-invasive neuroimaging provides the access to detect and visualize the functional and structural abnormalities in the brain, and has been used in clinical routine to provide decision support. Given longitudinal 4D neuroimaging data, we may able to capture the tissue changes over time, and further predict the development of the disease. This project aims to develop a 4D neuoimage computing system for tracking such tissue changes. It can also serve as a tool for clinical decision support, and for the research on progressive neurological disorders, such as dementia.

Required Programming Language: Python

Feature-Centric Content Analysis in Biomedical Images (12 or 18cp)
Great advances in biological tissue labeling and automated microscopic imaging have revolutionized how biologists visualize molecular, sub-cellular, cellular, and super-cellular structures and study their respective functions. How to interpret such image datasets in a quantitative and automatic way has become a major challenge in current computational biology. The essential methods of bioimage informatics involve generation, visualization, analysis and management. This project aims to develop automatic or semi-automatic approaches for content analysis in microscopic images, such as detection of certain cell structures, and tracing of cell changes over time. The studies will focus on computer vision algorithms and interactive framework development.

Preferred Programming Language: Matlab

Contextual Feature Representation for Image Pattern Classification (12 or 18cp)
Image pattern classification has a wide variety of applications, such as object detection and scene classification. The classification performance is largely dependent on the descriptiveness and discriminativeness of feature representation. Consequently, how to best model the complex visual features is crucial. Currently many different ways of image feature extraction have been proposed in the literature, yet their performance is still unsatisfactory and feature extraction remains a hot topic in computer vision. This project aims to study the various techniques of contextual feature representation and evaluate their effectiveness for different image applications.

Preferred Programming Language: Matlab

Intelligent 3D Single Neuron Reconstruction (18cp)
The single neuron reconstruction is one of the major domains in computational neuroscience, a frontier research area intersected with signal processing, computer vision, artificial intelligence and learning theory, applied mathematics, fundamental neuroscience and quantum physics. The 3D morphology of a neuron determines its connectivity, integration of synaptic inputs and cellular firing properties, and also changes dynamically with its activity and the state of the organism. Analyzing the three-dimensional shape of neurons in an unbiased way is critical to understanding how neurons function and developing applications to model neural circuitry. Such analysis can be enabled by reconstructing tree models from microscopic image stacks by manual tracing. However such manual process is tedious and hard to scale. This project aims to develop novel computational approaches for automatic 3D reconstruction of neuron models from noisy microscopic image stacks. Such methods would enable faster and more accurate neuron models to further accumulate the knowledge of single neuron functionality and neural network connectome.

Neuroimaging Computing for Early Detection of Dementia (18cp)
Dementia is one of the leading causes of disability in Australia, and the socioeconomic burden of dementia will be aggravated over the forthcoming decades as people live longer. So far, there is no cure for dementia, and current medical interventions may only halt or slow down the progression of the disease. Therefore, early detection of the dementia symptoms is the most important step in the management of the disease. Multi-modal neuroimaging has been increasingly used in the evaluation of patents with early dementia in the research setting, and shows great potential in mental health and clinical applications. The objective of this project is to design and develop novel neuroimaging computing models and methods to investigate pattern of dementia pathology with a focus on early detection of the disease.

Context Modeling for Medical Image Retrieval (18cp)
Content-based medical image retrieval is a valuable mechanism to assist patient diagnosis. Different from text-based search engines, similarity of images is evaluated based on comparison between visual features. Consequently, how to best encode the complex visual features in a comparable mathematic form is crucial. Different from the image retrieval techniques proposed for general imaging, in the medical domain, disease-specific contexts need to be modeled as the retrieval target. This project aims to study the various techniques of visual feature extraction and context modeling in medical imaging, and to develop new methodologies for content-based image retrieval of various medical applications.

Project supervised by Vera Chung

Video Object tracking for visiting applications (18cp)
This project is to study object tracking algorithms, including Particle Swarm Optimization (PSO), and apply the algorithms into video tracking in visiting applications. We have real-world animal video data captured from a zoo. You will implement the object tracking algorithms for these videos for animal tracking, give performance comparisons, and customized or improve the tracking results for this kind of applications. It is also possible to improve the tracking results by combining the video data and the simultaneous RFID data available (offline).

Requirements: good programming in C++ or Java.

Moving object detection for video surveillance (18cp)
In visual surveillance of both humans and vehicles, a video stream is processed to characterize the events of interest through the detection of moving objects in each frame. The majority of errors in higher-level tasks such as tracking are often due to false detection. This project will design and implement a new system to detect the moving objects in surveillance applications.

Requirements: good programming in C++ or Java.

Study of behaviour detection algorithm for objects in video streams (18cp)
Object motion plays an important role in determining the semantic understanding of videos. Object motion analysis and behavior understanding are the keys for developing intelligent systems in many domains such as visual surveillance and handwriting recognition. This project will survey and study object trajectory clustering which includes trajectory segmentation to identify a number of sub-trajectories in a trajectory, feature vector sequence formation, and visual word sequence creation.

Requirements: good programming in Python.

Deep Neural Networks for Object Detection (18cp)
Deep Neural Networks (DNN) exhibit major differences from traditional approaches for classification. They are deep architectures which have the capacity to learn more complex models than shallow ones. This expressively and robust training algorithms allow for learning powerful object representations. This project will exploit the power of DNN for the problem of object detection. This object detection problem address in this project is very challenging.

Requirements: good programming in Python.

Project supervised by Joachim Gudmundsson

Role-assignment of football players (12 or 18cp)
Co-supervised by Michael Horton
There is currently considerable interest in research for developing objective methods of analysing sports, including football (soccer).

Football analysis has practical applications in player evaluation for coaching and scouting; development of game and competition strategies; and also to enhance the viewing experience of televised matches. Until recently, the analysis of football matches was typically done manually or by using simple frequency analysis. However, such analysis was primarily concerned with what happened, and did not consider where, by whom or why. Recent innovations in image processing and object detection allow accurate spatio-temporal data on the players and ball to be captured.

In this project the aim is to implement a recent method to assign roles to players during the games. That is, as players constantly change roles during a match, we want to employ a "role-based"
representation instead of one based on player "identity". This, facilitates the possibility for a deeper analysis of the process of playing football matches.

Good programming skills and good algorithmic background are required.

Knowledge of basic aspects of machine learning would be very useful but not essential.

This is a 12cp project but can be converted to 18cp for students that can demonstrate a good background in algorithms and machine learning.

Dominant regions of football players (12 or 18cp)
Co-supervised by Michael Horton
There is currently considerable interest in research for developing objective methods of analysing sports, including football (soccer).

Football analysis has practical applications in player evaluation for coaching and scouting; development of game and competition strategies; and also to enhance the viewing experience of televised matches.

The aim of this project is to develop a fast algorithm to approximate the "dominant regions" of football players. A dominant region of a player is the region the player can reach before any other player. The general approach is to use a sampling method and a realistic model of a player’s movement.

Good programming skills and good algorithmic background are required.

This is a 12cp project but can be converted to 18cp only for students that can demonstrate a good background in algorithms.

Classifying attacks in football (18cp)
There is currently considerable interest in research for developing objective methods of analysing sports, including football (soccer).

Football analysis has practical applications in player evaluation for coaching and scouting; development of game and competition strategies; and also to enhance the viewing experience of televised matches.

This project looks at how to use tools from machine learning and computational geometry to classify different types of attacks ("pass-and-move", counter attacks, through balls,...) occurring in football.

Good programming skills and good algorithmic background are required.

Knowledge of basic aspects of machine learning would be very useful but not essential.

Classifying passes in football (18cp)
There is currently considerable interest in research for developing objective methods of analysing sports, including football (soccer).

Football analysis has practical applications in player evaluation for coaching and scouting; development of game and competition strategies; and also to enhance the viewing experience of televised matches.

This project looks at how to use tools from machine learning and computational geometry to classify different types of passes (direct, piercing, cross, backward,...) occurring in football.

Good programming skills and good algorithmic background are required.

Knowledge of basic aspects of machine learning would be very useful but not essential.

Project supervised by Bryn Jeffries

In-database motif detection of time series data (18cp)
Adapt a state-of-the-art motif detection algorithm to work within a PostgreSQL DBMS, handling issues such as paged memory and missing data, and investigate methods to optimise the algorithm through additional data structures. Apply to real datasets of, e.g., activity data from insomnia patients. This is honours or possibly strong Masters 18CP student project. Prerequisites are INFO3404 or INFO3504 (or equivalent), and student must be comfortable writing in C.

Feature extraction of speech task data (18cp)
Investigate automated segmentation algorithms to isolate key regions of speech recordings; implement candidate feature extraction algorithms to characterise the quality of speech for region; apply techniques such as Principle Component Analysis to identify the most distinctive features from extended wakefulness studies. This is honours or Masters 18CP student project. Student may need to work with C/C++ code so they should be comfortable with these languages.

App for actigraphy-assisted sleep diaries (12cp)
The importance of good sleep habits is becoming increasingly acknowledged amongst the general public, and an important step to improving better sleep habits is to keep an accurate log of bed-times, wake-times and quality of sleep. This can be a laborious and subjective process, but could be improved my making use of an actigraphy device such as a FitBit to make initial estimates that can be manually adjusted. In this project you would develop such an app. This project can be worked on by a group of students tackling several different aspects:

  1. Visualisation of sleep diary data
  2. Fusion of sleep diary and actigraphy data
  3. Knowledge discovery of sleep behaviour
  4. Client-server architecture for tracking sleep history

Projects supervised by Kevin Kuan

Mobile Web Beer Game (12cp)
The beer game is a classic experiential learning business simulation game to demonstrate a number of key principles of supply chain management. This project aims to implement this classic game on the mobile web platform that will be used in the classroom.

Requirements: Basic web programming skills; familiar with the beer game.

What Makes an Online Consumer Review Helpful? (12 or 18cp)
Description: Consumers increasingly rely on online product reviews in guiding purchases. This project aims to study different characteristics of online consumer reviews that are considered helpful to consumer decision making using secondary data from online review websites.

Requirements: Basic web programming skills; basic understanding in statistical analysis (e.g., regression); interest in text mining.

Emotional Contagion in Online Social Networks (12 or 18cp)
Description: The project is motivated by the recent controversial Facebook experiment on emotional contagion, in which Facebook manipulated the news feeds of nearly 700,000 users to examine if the emotion they expressed through messages on their news feeds influenced the emotion of other users as expressed in their subsequent posts (Kramer et al. 2014). To probe further into this Facebook experiment, this project aims to investigate factors affecting user posting behavior in online social networks using secondary data (e.g., from Twitter).

Skill requirements: Basic web programming skills; basic understanding in statistical analysis (e.g., regression); interest in text mining.

Effect of Information Overload in Online Decision Making (18cp)
Description: The negative consequences of information overload have been studied in a range of disciplines. Davis and Ganeshan (2009) showed that humans acquire and process relative more information under the threat of information unavailability. However, they are less satisfied with their decisions than those who acquire and process less information under no threat of information unavailability. This project extends the work on Davis and Ganeshan (2009) and investigates the impacts of information overload in the context of online decision making using controlled experiment.

Requirements: Basic web programming skills; basic understanding in statistical analysis (e.g., t-test, regression, etc.).

Project supervised by David Lowe

Augmented reality remotely-accessed labs (12 or 18cp)
Existing remote labs largely duplicate conventional experimental labs, but the computer interface provides an opportunity to enrich the experience of interacting with the equipment by using augmented reality approaches (imagine a magnetics experiment where the video image is overlayed to show the magnetic field lines). This project involves developing the software interfaces for an existing remote laboratory in order to provide an illustrative prototype. The prototype will demonstrate the benefits that can be achieved through the use of augmented reality technologies.

Architecture for collaborative remotely-accessed labs (12 or 18cp)
The leading remote labs software management system – Sahara – has been designed to be consistent with multi-student distributed collaboration, but this functionality has not yet been fully explored or implemented. This project will investigate extending Sahara to incorporate distributed student collaboration within an experiment session.

Remote control of physical systems: the role of context (12 or 18cp)
It is becoming increasingly common to use remote access to control physical systems. For example, researchers within the Faculty have been exploring remote and automomous control of Mars Rovers, mining equipment, teaching laboratory apparatus and fruit picking robots. This project will focus on the role of contextual information in supporting engagement and learning in these systems.

Simulation of chronic pain through control of pressure gloves (12 or 18cp)
(In conjunction with Prof Phil Poronnik, Biomedical Sciences)
Work is underway with the artist Eugenie Lee to develop an installation to inform the public about chronic pain. This project requires a glove that can be pumped up with viscous fluid under computer control (probably a good job for an Arduino). Essentially this is trying to emulate pressure etc on the hands to go along with the Oculus illusions…

Projects supervised by Josiah Poon

Information extraction from Discharge Summary (12cp)
Discharge summary is a report prepared by a clinician when a patient leaves a hospital or after a series of treatments. It details a patient's complaints, diagnostic findings, the prescribed therapies and the patient's response. It also contains recommendations on discharge. There is useful information that can contribute to data mining but, unfortunately, they are recorded in an unstructured manner. It will be help a lot if these can be extracted from these summaries. However, it is not an easy task because not only of spelling errors, but additional challenges like they contain a lot of special codes and abbreviation, as well as not following grammatical rules. The aim of this project is to develop a tool to extract desirable information from a set of discharge summaries from the Stroke Department for data analysis.

Project supervised by Bernhard Scholz

Build and Test Harness for Soufflé (12cp)
Soufflé is an open-source translator for a declarative Datalog-like programming language developed by Oracle Labs. The translator compiles a declarative specification into parallel C++ code. One of Soufflé's application is the specification of program analysis over large code bases with millions of lines of code. Soufflé aims at producing high-performance C++ code that can be compiled with the native compiler on the target machine.

Currently, the build system of Soufflé uses the standard UNIX automake tools, and there is no automated build system for building packages for contemporary operating systems. The first goal of this project is to implement a build harness for packages of a various operating systems including Debian/Redhat (ia32, amd64, arm), Mac OS X, and Windows. The second goal is to have an automated regression test system with an extended regression benchmark. In addition, a performance system using modern web-based tools such as Jenkins should be built. The third goal is to build an interactive web-page for Soufflé so that visitors of the web-page can trial Soufflé on the fly.

Requirements: familiarities with building open source projects in Linux, shell scripting, web-based productivity tools.

Language Bindings for Soufflé (12 or 18cp)
Soufflé is an open-source translator for a declarative Datalog-like programming language developed by Oracle Labs. The translator compiles a declarative specification into parallel C++ code. One of Soufflé's application is the specification of program analysis over large code bases with millions of lines of code. Soufflé aims at producing high-performance C++ code that can be compiled with the native compiler on the target machine.

Currently, Soufflé does have a C++ binding mechanism. However, Soufflé does not support language bindings to other languages. The aim of this project would be to implement multiple languages bindings for (1) Python, (2) Java, and (3) SQL systems. The challenges of this project will be to replicate relational data-structures in the host languages and implement high-performance interfaces such that information can be exchanged between the host language and Soufflé efficiently.

Requirements: good C++ knowledge, understanding of Python, Java, and the SQL language.

Profiler & Debugger for Soufflé (12 or 18cp)
Soufflé is an open-source translator for a declarative Datalog-like programming language developed by Oracle Labs. The translator compiles a declarative specification into parallel C++ code. One of Soufflé's application is the specification of program analysis over large code bases with millions of lines of code. Soufflé aims at producing high-performance C++ code that can be compiled with the native compiler on the target machine.

Currently, Soufflé has a rudimentary command line profiler for querying the performance of a Soufflé program. The aim of this project is to implement a graphical visualization of the profile information and extend the profiler with a simple query language that can retrieve information from the computed relations. The project should be implemented in C++ using QT Creator.

Requirements: good C++ knowledge

Project supervised by Masa Takatsuka

Remote Interactive 3D graphics (12 or 18cp)
This project aims to carry out the feasibility study on building interactive remove 3D graphics platform. It will be based on our new 3D encoding technology to deliver 3D data to remote locations over the existing network infrastructure. This project will build fully interactive system which allows remote users to interact with a server, which generates 3D data and deliver to the remote user.

Adaptive Self-Organising Map for streaming data (12 or 18cp)
This project focuses on developing a new learning algorithm for Self-Organising Map (Artificial Neural Network). The new algorithm will increase local elasticity on a part of Self-Organising Map so that it can adjust itself according to the newly presented data, which requires the SOM to carry out locally drastic changes. This work will allows us to develop a new SOM suited to visualise big streaming data in real time.

Multi-threaded 3D game engine (12 or 18cp)
Most 3D massive multiplayer online games are based on single thread architecture. This project aims to exploit multi core/cpu architecture to develop a new multi-threaded MMO game engine.

Projects supervised by Xiuying Wang, Jianlong Zhou and David Feng

Deep learning approach for medical image segmentation (18cp)
Recently the research community has seen great success using deep learning for image analysis tasks. For example, the Convolutional Neural Network (CNN) is one of the most widely used methods for object detection/recognition. This project will use a deep learning approach for medical image segmentation. The multi-layer convolution of CNN will be utilized for detection and segmentation of tumors/lesions from biomedical images. The project will include the investigation on the effective features for training and design of more feasible deep learning scheme for identification and segmentation of the disease.
The student is expected to have solid software development skills.

Assessment of quality of medical image segmentation with physiological measurements (18cp)
The quality of medical image segmentation is one of significant factors that affect user’s motivation to apply it in diagnostic decisions. This project aims to assess quality of medical image segmentations by analyzing user’s physiological responses such as galvanic skin response (GSR) and eye-tracker during reviewing segmentation results. The investigation of this project will result in various physiological patterns for different quality levels of medical image segmentations. Such results may contribute to the design of medical image segmentation assessment platforms, which automatically reveal segmentation qualities based on user’s physiological responses during reviewing segmentation results.

The student is expected to have solid software development skills.

Investigation of user trust on medical image segmentation (18cp)
Computer medical image segmentation is one of useful techniques for radiologists to understand medical images. However, there are often trust issues on segmentation results when using computer automatically generated segmented results in practical diagnostic decisions. This project aims to investigate factors that affect user’s trust on segmentation from different aspects, e.g. segmentation results presentation, segmentation methods, segmentation parameters and other aspects.

The student is expected to have solid software development skills.

Visual analytics of multi-stream life data for health monitoring (18cp)
Recent new technologies produce various kinds of data in human daily life. For example, data streams and digital traces emerging from personal sensors and devices, social media updates, and other electronic systems as well as even web search data are recorded in everyday life, which indicate people’s lifestyle and living environment. Lifestyle and environment are important determinants of individual health and wellbeing. This project aims to analyze these multi-stream life data with visual analytics techniques and transform these analyses to our understanding of human health and disease surveillance.
The student is expected to have solid software development skills.

Content-oriented deformable image registration (18cp)
Current image registration methods are either based on image intensity information or based on the extracted feature to derive deformation field. These registration schemas may introduce excessive deformation onto the local important regions, in which the local features are expected to be preserved. In our research, we will focus on derive deformation fields according to the image contents for more meaningful and sensible registration. We will use temporal medical images as testing data for our research.

The student is expected to have solid software development skills.

Projects supervised by Zhiyong Wang

Multimedia Data Summarization (12 or 18cp)
Multimedia data is becoming the biggest big data as technological advances have made it ever easier to produce multimedia content. For example, more than 300 hours video is uploaded to Youtbue every minute. While such wealthy multimedia data is valuable for deriving many insights, it has become extremely time consuming, if not possible, to watch through a large amount of video content. Multimedia data summarization is to produce a concise yet informative version of a given piece of multimedia content, which is highly demanded to assist human beings to discover new knowledge from massive rich multimedia data. This project is to advance this field by developing advanced video content analysis techniques and identifying new applications.

Predictive Analytics of Big Time Series Data (18cp)
Big time series data have been collected to derive insights in almost every field, such as the clicking/view behaviour of users on social media sites, electricity usage of every household in utility consumption, traffic flow in transportation, to name a few. Being able to predict future state of an event is of great importance for effective planning. For example, social media sites such as Youtube will be able to better distribute popular video content to their caching servers in advance so that users can start watching the videos with minimal delay. This project is to investigate existing algorithms and develop advanced analytic algorithms for higher prediction accuracy.

Human Motion Analysis, Modeling, Animation, and Synthesis (12 or 18cp)
People are the focus in most activities; hence investigating human motion has been driven by a wide range of applications such as visual surveillance, 3D animation, novel human computer interaction, sports, and medical diagnosis and treatment. This project is to address a number of challenge issues of this area in realistic scenarios, including human tracking, motion detection, recognition, modeling, animation, and synthesis. Students will gain comprehensive knowledge in computer vision (e.g. object segmentation and tracking, and action/event detection and recognition), 3D modeling, computer graphics, and machine learning.

Intelligent multi-video editing (12cp)
While it has been increasingly easier to take videos of the real world, it is still very challenging to create a high quality video from multiple videos taken by multiple devices. One key reason is that different settings of multiple imaging devices will result in subtle differences which will affect view experiences. This project is to investigate the challenging issues in multi-video editing and develop intelligent tools to streamline the editing process.

Food recognition (12cp)
Recent years have witnessed significant progress of object recognition in computer vison domain, due to the advancement of machine learning techniques such as deep learning. This project is to investigate the state-of-the-art object recognition techniques and develop a computer system to recognize specific food.

View synthesis for VR/AR (12cp)
Virtual Reality/Augment Reality techniques are able to enhance human perception with immense experiences. One of the key technical components is to generate realistic visual content. This project is to develop techniques for bring better visual experiences to VR/AR users.

Project supervised by Bing Zhou

Computing and visualising transcription factor interaction networks (18cp)
The gene expression of the cells is regulated by the binding and interaction of various transcription factors (TFs). With the enormous amount of ChIP-sequencing (ChIP-seq) data generated for TFs by projects such as ENCODE and modENCODE, it is possible to reconstruct the TF interaction networks by correlating and modelling the binding profiles of TFs.

We will be developing high-performance computing algorithm to dynamically computing correlations among a large set of TF binding profiles. The correlation matrix will then be used to reconstruct the TF interaction networks. We also aim to develop a graphical user interface to display the TF interaction networks by methods such as clustering.

In this project the students will get involved in 1: Algorithm design, implementation and testing on multicore computers and clusters of PCs; and 2: An interactive graphical user interface design and implementation.

Requirements: good programming skill (essential) and experience in graphical user interface development (desirable).

Project supervised by Ying Zhou

Visualizing Wikipedia Co-authorship Network (12cp)
Wikipedia is maintained by mass collaborative efforts. Each wikipage is coauthored by many editors. The edit history of wikipages show lots of interesting patterns.

Such patterns are used by bot algorithms to detect potential vandals, to flag if certain page content is contentious and many other useful feature.

This project aims to extract co-author information from the talk and edit history pages and to visualise the co-authorship network along the timeline with various degree of details.

Required skill: Python, JavaScript, Graph Processing

Project supervised by Albert Zomaya and MohamadReza Hoseiny

Energy-Efficient Qos-Aware Dynamic Workload Consolidation in Virtualized Cloud Datacentre (12 or 18cp)
Workload consolidation (either using virtualization techniques or container-based methods) has attracted lots of attention in big cloud datacentres nowadays. In this project we try to find a best trade-off to reach an efficient energy consumption among shared resources as well as optimized performance level of running collocated applications in a modern data centre which normally contains thousands of multi/many core boxes severed as a future cloud computing infrastructure. Past researches showed that performance interference due to shared resources across co-located virtual machines make today's cloud computing paradigm inadequate for performance-sensitive applications while much expensive than necessary for the others. To satisfy quality of service demanded by customers, cloud providers have to deploy VMs in more PMs (than actually is needed) to achieve a better resource isolation which in turn adversely consumes higher level of energy. Hence, a challenging problem for providers is identifying (and managing) performance interference between the VMs that are co-located at any given PM. In this project, we aim to propose consolidation algorithms to reduce energy consumption of a given data centre while avoiding the system performance degradation in the same time with a focus on the impact of shared resources such as last level cash (LLC), and so on.

Building a flexible event-driven framework for cloud based applications (12 or 18cp)
Unlike traditional application architectures which are too static, inflexible and cumbersome to implement, the modern 'event-driven paradigm' allow applications to be modified rapidly and to quickly respond to errors and exceptional conditions that disrupt conventional processes. Event-driven application systems help enterprises react quickly and precisely to rapidly changing conditions. Allowing the transmission of events among loosely coupled and highly-distributed software services, EDA is well suited for cloud based applications focused on scalability and flexibility features. In this project, we concentrate on the concept of event driven application frameworks as we believe that it is the underlying key factor that will enable revolutionary improvements in large scale distributed business systems build on hybrid cloud model in near future. We try to design and implement an application framework which is intended to reduce the coupling between application layers, and achieve a better distribution of responsibilities between different services. Candidate for this project must have strong passion and knowledge in one (or more) of the functional programming language such as Scala (Akka esp.), Erlang, Clojure, Haskell and OO programming languages such as Java, Ruby, or Python. In addition, strong knowledge of concurrency, FP, Actor-based systems, object-oriented design, and software design patterns and principles is a must.

Designing the optimal data model of large-scale applications in Google App Engine (12 or 18cp)
Google App Engine is a famous infrastructure for building highly scalable web applications. Data objects, also known as entities, are saved in the App Engine datastore which is build on top of Google's NoSQL database, Bigtable. Every attempt to modify an entity must happen in the context of a transaction. In fact, transaction ensures that all of the operations it contains are applied as a atomic operation to keep data consistency. However, the big issue is that Google App Engine's datastore has limited support for transactions. In particular, to guarantee that updates to two or more entities are atomic, these entities must be in the same entity group, or in a maximum of five entity groups, known as cross-group (XG) transaction. On the other hand, entity groups should keep small number of entities to avoid possible contention during the lifetime of a running application. This means that there are two conflict objectives of having consistency among the data and reaching peak performance of system. In this project, we try to design some algorithms to advise how a business should organise its data when uses Google App Engine datastore. In particular, we will create a software system to make an optimal decision for reaching a tradeoff between the level of consistency and performance regarding several other factors such as the total budget, the limits of quota, and performance characteristics. Students who desire to work on this project must have strong background on database systems, as well as Java and Python programming languages.

Optimal scheduling of bag-of-task application in hybrid cloud environment in the presence of heavy-tailed tasks with bursty traffic arrival pattern (12 or 18cp)
The aim of this project is to investigate the optimal scheduling of Bag-of-Task applications in hybrid cloud environment (comprising of both private and public resources) in the presence of heavy-tailed and bursty traffic. This traffic model is considerably different from traditional arrival traffic patterns which can be modelled using Poisson or Markov-modulated processes. However, empirical evidence supported by deep analysis on workload of todays' applications in several Grid and Cloud environments showed that the traffic of tasks' arrival is intrinsically bursty, and exhibits correlations over longer time scales than the traditional models. In such scenarios that involves application with heavy-tailed running time tasks, there are relatively few studies on the problem of scheduling of tasks on hybrid cloud resources. One of the key considerations in the design of such a scheduling policy is the concept of Pareto- optimality of multiple objective functions such as total incurring cost and/or makespan. The output of this project will include an implementation of suggested scheduling algorithm in the real cloud environment with deep analysis of several well-known workload patterns.

Running Workflow application (Montage as a case study) in Cloud (12 or 18cp)
Amazon EC2 and S3 resources offer computational and storages elements can be used on-demand by compute and data-intensive applications. I am working with my colleagues at the University of Sydney to address problem of how we could implement a workflow application cost-effective in the current cloud resources. We pick up Montage, an I/O bound astronomy application, to compare its running performance on the Amazon EC2 cloud and Sydney Uni. High performance cluster.

PANDA: a PAreto Near-optimal Deterministic Algorithm for Scaling Out Large-Scale Bag-of-Tasks Applications Across Multiple Clouds (12 or 18cp)
Large-scale Bag-of-Tasks (BoT) applications are characterized by their massively parallel, yet independent operations. The use of resources in public clouds to dynamically expand a private cloud might be an appealing alternative to cope with such massive parallelism. To fully realize the benefit of this 'multi-cloud' deployment, cost efficiency must be thoroughly studied and incorporated into scheduling and resource allocation strategies. In this research project, we are going to build a system to address the problem of running a large number of tasks (BoT application) on resources in a multiple cloud environment capturing the trade-off between cost and time/performance. Specific contributions in this work are (1) the heterogeneity of both resources and tasks is taken into account in our algorithms, (2) cost and performance of running jobs across multiple clouds are effectively modelled in our objective function, (3) our system guarantees that the quality of task assignment is optimal or at least near-optimal, (4) The performance and practicality of our system have been thoroughly evaluated in terms of both performance and cost with experiments in Amazon EC2 environments using a real-world BoT application.

Project supervised by Albert Zomaya and Wei Li

Addressing Interoperability in the Internet of Things (12 or 18cp)
The advances on electronic devices, wireless communications, RFID technology, and the explosive growth of the World Wide Web contributed to leverage the development of the Internet of Things (IoT) paradigm. In the IoT vision, every object on Earth can be found and used via Internet. Interoperability is one of the major focuses in IoT to provide the foundation of interaction between human-to-machine(H2M) and machine-to-machine(M2M). There are different perspectives of interoperability to be addressed in IoT systems, such as technical, syntactic and semantic. Technical interoperability concerns the integration, mainly through communications protocols, of the myriad of heterogeneous physical devices and systems that support M2M interactions. Syntactic interoperability focuses on the formats of the data sent in the payload of the messages exchanged by the devices. Such data needs to conform to a well-defined formatting and encoding so they can be consumed by all stakeholders involved in the communication. At a higher level, the semantic Interoperability plays the role of ensuring that the actors involved in the communication have a common understanding not only of the format but of the content of the exchanged messages.

The project is aimed at improving the provision of technical and syntactic interoperability by implementing support to COAP (Constrained Application Protocol). COAP is a protocol developed to allow simple electronic devices (with limited computational and energy capabilities) to be able to communicate over the Internet while taking into account four main factors: (i) battery, (ii) latency, (iii) transmission capability, and (iv) mobility. In the project we intend to add support for semantic interoperability by adopting ontologies specifically tailored for IoT environments.

Computation Offload in Fog Computing (12 or 18cp)
Fog computing is a newly introduced concept as an extension of cloud computing to provide computing, storage and networking services in between end devices and traditional cloud computing data centres. It aims at providing the QoS requirements, such as: mobility support, geographical distribution, location awareness and low latency that required by the internet of things (IoT)’ applications but the traditional cloud computing is failed to address. To offload the computation workload in fog devices can overcome the resource constraints on the edge devices so as to saving storage and increasing battery lifetime, and more importantly providing better QoS requirements to IoT applications.

This projects aims at dealing with the dynamic of computation offload in fog devices. The dynamic has three fold 1) the fog device accessing is highly dynamic, 2) the fog device availability is highly dynamic, and 3) the fog device resource is highly dynamic. With such dynamics, questions like what kind of granularity to choose for offloading at different hierarchy of fog and cloud, how to dynamically partition application to offload on fog and cloud and how to make offloading decisions to adapt dynamic changes in network, fog devices and resources are eager to answer.

Handling Big Data-Streams in the Internet of Things (12 or 18cp)
In the Internet of Things (IoT), with billions of nodes capable of gathering data and generating information, the availability of efficient and scalable mechanisms for collecting, processing, and storing data is crucial. The number of data sources, on one side, and the subsequent frequency of incoming data, on the other side, create a new need for Cloud architectures to handle such massive flows of information, thus shifting the Big Data paradigm to the Big Stream paradigm. Moreover, the processing and storage functions implemented by remote Cloud-based collectors are the enablers for their core business, which involves providing services based on the collected/processed data to external consumers. Several relevant IoT scenarios, (such as industrial automation, transportation, networks of sensors and actuators), require real-time/predictable latency and could even change their requirements (e.g., in terms of data sources) dynamically and abruptly. Big Stream-oriented systems could react effectively to changes and provide smart behaviour for allocating resources, thus implementing scalable and cost-effective Cloud services. Dynamism and real-time requirements are another reason why Big Data approaches, due to their intrinsic inertia (i.e., Big Data typically works with batch-based processing), are not suitable for many IoT scenarios. The Big Stream paradigm allows performing real-time and ad-hoc processing in order to link incoming streams of data to consumers, with a high degree of scalability, fine-grained and dynamic configuration, and management of heterogeneous data formats. In brief, while both Big Data and Big Stream deal with massive amounts of data, the former focuses on the analysis of data, while the latter focuses on the management of flows of data. This characteristic has an impact also on the data that are considered relevant to consumer applications. For instance, while for Big Data applications it is important to keep all sensed data in order to be able to perform any required computation, Big Stream applications might decide to perform data aggregation or pruning in order to minimize the latency in conveying the results of computation to consumers, with no need for persistence.

The goal of this project is to investigate data centric architectures and algorithms to efficiently manage Big Stream data flows in the Internet of Things.

Task Scheduling Algorithms for the Internet of Things (12 or 18cp)
Applications for the Internet of Things are typically composed of several tasks or services, executed by devices and other Internet available resources, including resources hosted in a cloud platform. The complete execution of an application in such environments is a distributed, collaborative process. To enable collaborative processing of applications, the following problems must be solved: (i) assigning tasks to devices (and other physical or virtual resources), (ii) determining the execution sequence of tasks, and (iii) scheduling communication between involved devices and resources. Unlike traditional task scheduling, with the involvement of sensors and actuators, the types of tasks in IoT applications are more than computation and communication. Different allocation of these tasks on the nodes consumes different amounts of resources from the limited devices and provides different quality of service (QoS) to the applications. Although QoS management in traditional distributed and networked system is a well-studied topic, in IoT this is still a poorly investigated subject and the definition of QoS in IoT is still not clear. The traditional QoS attributes such as throughput, delay, or jitter are not suitable in IoT, where additional attributes are concerned, such as for instance the information accuracy (that is qualified with the probability that an accuracy can be reached), and the network resources required. Therefore, a task scheduling algorithm should handle the efficient execution of a large number of applications in the IoT infrastructure considering multiple types of resources and different QoS parameters, specific to IoT environments. Moreover, such an algorithm must consider situations in which multiple applications can perform common tasks (such as the sensing of the same physical variable), that do not need to be performed several times by the devices and physical resources, thereby saving energy. Still, there may be applications with higher priorities in relation to the required response time (e.g., critical time applications) or in relation to the amount of resources provided to them (e.g., bandwidth, sensing coverage), compared to others that are sharing the IoT infrastructure and therefore these priorities must be respected when allocating tasks.

This project aim to study existing task scheduling algorithms, especially those for cloud computing and wireless sensor network environments, and to propose a new one, specifically tailored for IoT environments.

Virtualization of Wireless Sensor Networks (12 or 18cp)
In the last few years, we have witnessed the emergence of a new paradigm, called the Cloud of Things (CoT). The CoT paradigm emerged from the combination of the paradigms of Cloud Computing and Internet of Things (IoT). Essentially, in the CoT paradigm, the Cloud acts as an intermediate layer between smart things and applications. Such an intermediate layer hides the complexity of smart things necessary to implement applications and allows that IoT systems can benefit from the virtually unlimited resources of a Cloud to store the huge amount of data produced by the interconnected device and to implement sophisticated processing and data analytics. Similarly, the Cloud can benefit from the IoT by extending its scope to deal with real world objects (smart things) in a distributed and dynamic way. Among the highly heterogeneous set of smart things composing CoT environments, smart sensors and actuators play an important role. Smart sensors and actuators present some specific features that need to be considered when integrating them with the Cloud, namely: (i) such devices have a lower degree of heterogeneity and mobility in comparison to some devices typical of IoT, such as wearable sensors or field operation devices; and (ii) such devices are more resource constrained, specifically in terms of the available energy for operating. The potential advantages brought by wireless sensor and actuator networks (WSAN) along with the specific features of such devices have motivated the emergence of the Cloud of Sensors (CoS) paradigm [1], as a type of ecosystem within the broader domain of CoT.

A CoS is composed of virtual nodes built on top of physical WSAN nodes, and provides support to several applications which, in turn, may require access to functionalities at the Infrastructure as a Service (IaaS), Platform as a Service (PaaS), and Software as a Service (SaaS) levels. Application owners can automatically and dynamically provision such virtual nodes on the basis of applications requirements. In this sense, the CoS infrastructures are built centred on the concept of wireless sensor network virtualization, which is expected to provide a clean decoupling between services (required by applications) and infrastructure (encompassing sensors, actuators and the cloud).

In this context, the main goal of this project is to propose a new model for wireless sensor network virtualization. By pursuing this we aim to answer two research questions: (i) How to virtualize devices (sensors, actuators and other physical objects) and networks of such devices, and (ii) How to properly separate the responsibilities between the components of a CoS (devices, applications and cloud platform). For doing so, we will initially investigate existing conceptual models for the integration of physical devices with the cloud and propose (or adopt) one the fits the specific requirements of CoS. The conceptual layers of the proposed model will be defined, their responsibilities and the interactions among the envisioned components. The proposed model will promote loose coupling between the layers in order to cope with the system scalability, and to promote a clear separation of responsibilities between the layers, isolating issues related to the (sensing and communication) infrastructure from issues linked to the application.