Skip to main content

Deakin Research

Deakin Research

Contact us


+61 3 5227 1266


Centre for Pattern Recognition and Data Analytics
School of Information Technology
Deakin University
Locked Bag 20000

What we do

In theory

Our methods are grounded in statistical machine learning and pattern recognition. This includes probabilistic graphical models, such as Markov models, hierarchical hidden Markov models and conditional random fields.

We are now focused on:

  • Probabilistic models
    Our work includes probabilistic graphical models, such as Markov models, hierarchical hidden Markov models, conditional random fields and more recently, Bayesian non-parametric models.

    We ask two questions:
    • What are the models that we must construct to be computationally scalable in big-data problems?
    • How do we enhance the expressiveness of such models, what are the representations and can we develop appropriate inference and learning?

  • Compressed sensing and sparsity modeling
    What are the sparse, robust models we can construct for big data?

What are the domains that inspire new models?

Large-Scale Surveillance

We are interested in the problems that arise when we examine hundreds of cameras. This big data problem opens up many new questions: Given limited operator capacity, how can we select which camera an operator should look at? How do we select models and parameters for heterogeneous stream data - for example, video feeds in varying environmental conditions? How do we evaluate algorithms in big data - training sets may exceeds two weeks of video data, and no realistic ground truth can be obtained from operators. How do we compare algorithms?

What we have done so far

Focusing on issues of large scale surveillance we have developed new techniques to model "normal" data from static video cameras. This allows us to detect real time "abnormal events" and thus enable operators to focus on the 1% of events in a video feed. Our algorithms drive the start-up iCetana's innovative anomaly detection software. The software uses ideas from Compressed Sensing to enable simultaneous surveillance of many cameras deployed in diverse settings. A local city council has used our algorithms to detect loitering, anti-social behaviour and traffic violations. For more information see ICETANA. The technology was: Winner, The Broadband Innovation Award, Tech23, 2010; Winner, 2011 WA Innovator of the Year..

Social Media Analysis

We study patterns in social media data; who is talking to who? What are they talking about? What is their sentiment? We discover latent community structures, their topics and trends. We explore not only text in blogs but also connectivity through comments. For personal media management it could be argued that social context contains the set of conceptions most often brought to bear on their contents: Where was this? Who is that? What activity were we all doing there? Who do I want to share this with? This project aims to use signals obtained from unobtrusive data sources available with today's devices, such as location information (e.g. GPS), to extract socially meaningful indices for personal media.

Social spheres are locations of significance, such as Home, Work and Recreational Area. Social networks of social ties capture the dynamic web of relationships in which we are embedded. Social rhythms arise as patterns across these former elements and allow even finer resolution categorization of activities that occur within them in space and time. All of this information can then be used, at the very least, to provide new ways to index and interact with our own, or others', media collections. Initial work in this area focused on discovering the social spheres Work, Home and Other from raw, noisy GPS traces of everyday life, and used this information together with the presence of known persons to provide a novel personal media exploration environment called Socio-Graph. The interface aimed to allow simple search and filtering on the concepts most people innately bring to their personal media: social. We do this project jointly with our colleagues at IMPCA.

Tools for Early Intervention in Autistic Children

There is a growing gap between the number of children with autism requiring early intervention and available therapy. In this project we seek to produce the following outcomes:

  1. Flexible frameworks for stimulus presentation and recording, for early intervention in social and cognitive/visual areas,
  2. Open source infrastructures to leverage content and metadata from social media, constructing foundations for reusable content in early intervention,
  3. Assistive Technologies for finding appropriate information from forums and blogs, focusing on autism,
  4. Early warning systems of mental wellbeing, measuring affect in text-based social media (forums, blogs, etc.) and appropriate triggers for social support.
  5. New tools for assistive support for communication and social function.

Our partners are: Autism West, Gateways, Barwon Health.
We do this project jointly with our colleagues at IMPCA.

What we have done so far?

We present a portable platform for pervasive delivery of early intervention therapy using multi-touch interfaces and principled ways to deliver stimuli of increasing complexity and adapt to a child's performance. Our implementation weaves Natural Environment Tasks with iPad tasks, facilitating a learning platform that integrates early intervention in the child's daily life. The system's construction of stimulus complexity relative to task is evaluated by therapists, together with field trials for evaluating both the integrity of the instructional design and goal of stimulus presentation and adjustment relative to performance for learning tasks. We show overwhelmingly positive results across all our stakeholders, children, parents and therapists. Our results have implications for other early learning fields that require principled ways to construct lessons across skills and adjust stimuli relative to performance. See TOBYPLAYPAD.

The technology won the 2011 Curtin University Commercialisation award.

Health Analytics

With our partner Barwon Health we are embarking on questions that arise in examination of large, disparate and multimodal hospital data sets. Can we impact and inform the formation of dynamic health intervention and improved safety and care through?

  1. Predicting hospital visitation patterns of patients with chronic disease,
  2. Detection of sub-populations that have coherent patterns of disease, causal factors, and typical medical responses.
  3. Identify data driven characteristics of chronic patients to enable personalized care plans.
  4. Detect indicators that serve as early warning of chronic disease.
  5. Monitor key factors that deliver value to patients in the areas of care experiences, care coordination and patient safety.

Deakin University acknowledges the traditional land owners of present campus sites.

27th February 2015