December 2008

• Wednesday, 17 December 2008, 1-2pm*
  Host: Punarjay Chakravarty (Faculty of Engineering)
  Title: People Tracking from a Moving Panoramic Camera
  Abstract: This talk presents a method of detecting and tracking people from a moving panoramic camera, such as one mounted on a mobile robot. People's appearances in panoramic images are trained using the Adaboost algorithm working on an extended set of Haar-like features. A multiple target tracker is presented that is able to track people over time based on the Haar detections, weed out the false alarms, and maintain tracks despite missed detections. The tracker is able to maintain track IDs through crossing tracks without the requirement of depth information.


• Friday, 12 December 2008, 1-2pm*
  Host: Greg Paperin
  Title: Dual Phase Evolution and Self-Organisation in Networks
  Abstract: Complex adaptive and evolutionary systems exhibit a sustained diversity, far-from-equilibrium dynamics, and permanent novelty and adaptation in the absence of a global controller. Previous work shows that many such systems can be represented as networks of interacting components. These networks are typified by certain complex topologies. Insights into the processes behind the emergence of complex network structures and into the effects of such structures are necessary for an understanding of properties that characterise adaptive and evolutionary systems. Dual Phase Evolution (DPE) is a widespread natural process in which networks underlying complex systems adapt and self-organise by switching alternately between two phases: a phase of global interactions and a phase of local interactions. Each phase is characterised by specific global connectivity and interaction patterns. Here, I present ongoing work on DPE in complex evolutionary systems. I show how DPE processes can give rise to a wide variety of complex network topologies. In particular, this includes the emergence of scale-free degree distributions in fixed-size networks, as well as modular structures. I also show how DPE can be responsible to the continuous novelty observed in many natural and artificial evolutionary systems.


• Wednesday, 3 December 2008, 1-2pm*
  Host: Sk. Mohammad Rokonuzzaman
  Title: Multiple Radio Channels and Directional Antennas in Suburban Ad Hoc Networks
  Abstract: The Suburban Ad Hoc Network (SAHN) is a cooperative ad hoc wireless mesh network. Nodes are owned and operated by end-users without reliance on central infrastructure. It provides symmetrical bandwidth allowing peer-to-peer services and distributed servers. We minimize the use of scarce unlicensed RF spectrum supported by Smart Antenna technology. RF interference in such networks and techniques and strategies to reduce it are examined. Traffic is spread across multiple frequency channels and multiple directional beams to achieve improved spatial re-use. We focus on the control of Smart Antennas rather than their design. By dynamically adjusting our network topology using Smart Antennas and dynamically re-routing current communications we optimize the network for its current traffic needs.

November 2008

• Wednesday, 26 November 2008, 1-2pm*
  Host: Geoff Webb
  Title: MultiStrategy Ensemble Learning, Ensembles of Bayesian Classifiers, and the Problem of False Discoveries
  Abstract: This talk covers an ensemble of my research contributions that I believe are likely to resonate with a current audience. Ensemble Learning combines the predictions of multiple classifiers to enhance accuracy relative to any individual classifier. I will show that combining established ensemble learning techniques further enhances accuracy without computational overhead. Naive Bayes is a popular approach to classification learning due to its computational efficiency, strong theoretical foundation and its capacity to predict probabilities rather than just the most probable outcome. I will present a simple extension that creates an ensemble of naive-Bayes like classifiers, improving naive Bayes' accuracy without undue computational burden. Finally, I will discuss false discoveries, a problem that plagues many modern pattern discovery systems. Quite simply, many state-of-the-art approaches to pattern discovery are prone to 'discover' patterns that do not exist. I will explain why this is so and discuss approaches to overcome the problem.


• Tuesday, 18 November 2008, 1-2pm*
  Hosts: Kerri Morgan and Amiza Amir
  Title: Colouring, Cooking, and Factorising
  Abstract: Some students have expressed an interest in making their own rocky road. Amiza and I will demonstrate how simple it is to make this sweet. Samples will be provided. If time permits I will also give a talk on "Chromatic Factorisation of Graphs". The chromatic polynomial gives the number of proper colourings of a graph. We define the concept of chromatic factorisation. It is known that every clique-separable graph has a chromatic factorisation. However, we identify a number of graphs that have chromatic factorisations but are not clique-separable. We then show that every graph with chromatic number at least three is a chromatic factor of some chromatic factorisation.


• Wednesday, 12 November 2008, 1-2pm*
  Hosts: Enes Makalic and Daniel Schmidt
  Title: Applications of Information Theory to Statistics
  Abstract: This brief informal talk covers a basic introduction to the theory of coding (i.e., Kraft's inequality, entropy, etc.). Methods to be discussed include: Akaike's Information Criterion (AIC), Bayesian Information Criterion (BIC), Minimum Message Length (MML) and Minimum Description Length (MDL). Several applications of information theory to statistical problems will also be discussed.


• Wednesday, 5 November 2008, 1-2pm*
  Host: Nghia Ho (Faculty of Engineering)
  Title: Vision-Based Global Localisation (Where Am I?)
  Abstract: This talk demonstrates how a robot can perform global localisation (where am I?) using a panoramic mirror in conjunction with a rich, previously acquired 3D model of the environment and a particle filter for localisation. The robot can estimate its position and orientation given a sequence of images taken from the test environment. This talk will mainly be a summary of what I have done for the past 3.5 years of my PhD and will include videos.

October 2008

• Wednesday, 29 October 2008, 1-2pm*
  Host: Anang Hudaya Muhamad Amin
  Title: A Divide-and-Distribute Approach to Single-Cycle Learning of HGN Networks
  Abstract: The Distributed Hierarchical Graph Neuron (DHGN) is a novel single-cycle learning, distributed pattern recognition algorithm, which reduces the computational complexity of existing pattern recognition algorithms by dividing and distributing the recognition process into smaller clusters. The proposed approach extends the single-cycle pattern recognition capability of the hierarchical graph neuron (HGN) for wireless sensor networks into the more generic framework of computational grids. The computational complexity of the hierarchical pattern recognition scheme is significantly reduced and the accuracy is improved. The single-cycle learning capability, which develops within the HGN, shows better noisy pattern recognition accuracy when the size of the clusters is adapted to pattern data. In this presentation, the author will discuss the effect of dividing and distributing simple pattern recognition processes within a computational network, and how it affects the accuracy and efficiency of the proposed pattern recognition scheme.


• Wednesday, 22 October 2008, 1-2pm*
  Host: Raja Azlina Raja Mahmood
  Title: Detecting Attack Patterns Using an Efficient and Low-Computational Pattern Recognition Scheme
  Abstract: The proposed pattern recognition scheme employs a Distributed Hierarchical Graph Neuron (DHGN) technique. The DHGN is a new form of neural network which consists of a hierarchical graph-based representation of input patterns and adopts a one-cycle learning process. We have proven the technique to be efficient; real-time and accurate, and offer low computational complexity, in a few studies. The author will discuss the proposed technique and its application in detecting attack patterns in Mobile Ad Hoc Networks (MANETs).


• Wednesday, 15 October 2008, 1-2pm*
  Host: Benjamin Porter
  This lunch is not a seminar but a chance to kick back and watch some fun 3d short films. Come and be inspired by the fantastic intersection of computer science and art!


• Wednesday, 8 October 2008, 1-2pm*
  Cross-faculty introduction session with open discussion
  Moderator: Mauro Bampo
  Discussion topic: Computational Intelligence from Your Perspectives
  Abstract: Artificial intelligence has been an active research field for the last decades, and involves many fields such as computer science, cognitive science, operations research, control engineering, philosophy, neurobiology, evolutionary biology, and linguistics. Yet, there is still no universally accepted definition of intelligence, with many questions unanswered and being topics for debates. These questions include: Can machines behave intelligently? Will they match or exceed human intelligence? What are the limits of a machine's capabilities? Is an intelligent agent really intelligent? Can it think and have a mind, or can it only act like it thinks and has a mind? Have we succeeded yet? After all, what exactly are we trying to do? In this seminar, we will discuss the notion of computational intelligence in the context of our diverse research areas.

September 2008

• Tuesday, 30 September 2008, 1-2pm*
  Host: Cameron McCormack
  Title: Authoring Adaptive Diagrams
  Abstract: The web and digital media requires intelligent, adaptive documents whose appearance and content adapts to the viewing context and which support user interaction. While previous research has focussed on textual and multimedia content, this is also true for diagrammatic content. We have designed and implemented an authoring tool which supports the construction of adaptive diagrams. Adaptive layout behaviour is specified by using constraint-based placement tools as well as by allowing the author to specify more radical layout changes using alternate layout configurations. As well as specifying alternate layouts, the author can specify alternate representations for an object, alternate styles and alternate textual content. The resulting space of different versions of the diagram is the cross product of these different alternatives. At display time the version is constructed dynamically, taking into account the author specified preference order on the alternatives, current viewing environment, and user interaction.


• Wednesday, 24 September 2008, 1-2pm*
  Host: Michael Tyson
  Title: Developing Radio Propagation Environment Awareness (RPEA) for Mobile Ad Hoc Networks
  Abstract: RF Propagation Environment Aware (RPEA) mobile ad hoc networks are able to store and exploit local radio propagation information to predict future link quality and lifetime. This provides for optimisation of network management operation, and facilitates Quality of Service support. A model of the radio propagation geometry of the local environment is developed and maintained by observing signal loss measurements over time, while operating in the area of interest. This seminar will discuss factors related to the establishment of a radio propagation geometry model from observed signal losses.


• Tuesday, 16 September 2008, 1-2pm*
  Host: Dhananjay Thiruvady
  Title: Strip Packing with Hybrid ACO – Placement Order is Learnable
  Abstract: In this talk I will present a hybrid ant colony optimization algorithm for the strip packing problem. In this problem a fixed set of rectangular items of fixed sizes have to be placed on a strip of fixed width and infinite height without overlaps and with the objective to minimize the height used. To place items on the strip the bottom-left fill (BLF) heuristic is used. This study shows that learning item order with ACO combined with BLF placement is an effective algorithm for strip packing.


• Wednesday, 10 September 2008, 1-2pm*
  Host: Jenny Kashmirian
  Title: Programming Biology
  Abstract: The paradigm of the emerging field "Synthetic Biology" is programming the cell. Proponents are working on a tool set called biobricks which will simplify cell programming. They envisage two professions will emerge from this field, the designers (programmers) and the implementors. The designers will work at the abstract level of biobricks and the implementors will create the organisms containing the biobrick components. The success of this field will be determined in part on the ability to predict emergent effects at two levels: (1) cell behaviour – the effect of combinations of biobricks on the internal cell mechanism and therefore the overall cell behaviour; and (2) population behaviour – how individual cell behaviour effects overall population behaviour. The question is posed: What role should computer scientists play in this emerging field?


• Wednesday, 3 September 2008, 1-2pm*
  Hosts: Mauro Bampo and Michael Wybrow (FIT Casual Robotics Group)
  Title: Lego Mindstorms Robots for Open Day 2008
  Abstract: The FIT Casual Robotics Group is a group of students and staff members who mess around with robotics projects in their spare time. In this talk we describe and demo several Lego Mindstorms robots we prepared for Monash University Open Day 2008 – a feeding automaton, a catapult and a maze navigating robot. We describe our experience with the Lego Mindstorms robotics kits and discuss the issues related to programming these robots.

August 2008

• Wednesday, 27 August 2008, 1-2pm*
  Host: Arun Mani
  Title: An Inequality for Counting Forests in Graphs
  Abstract: Counting the number of forests in a graph is known to be a #P-hard for most graph classes. In this talk we discuss an inequality that gives an upper bound for this number using graph decompositions.


• Tuesday, 19 August 2008, 1-2pm*
  Host: Cagatay Goncu
  Title: Automatic Generation of Accessible Diagram by Semantic Preserving Adaptation
  Abstract: With this research, we are investigating to develop techniques for presenting diagrams to blind people in alternative and more effective ways through different modalities such as tactile, audio, and tactile/audio.


• Tuesday, 12 August 2008, 1-2pm*
  Constituting event
  Host: Daniel Schmidt
  Title: Parameter Estimation of Autoregressive Models using the Minimum Message Length Principle
  Abstract: This seminar discussed a new parameter estimation procedure for autoregressive models derived from the Minimum Message Length (MML) principle. The resulting MML estimator tends to shrink the model poles towards the origin, leading to improved multiple step squared prediction errors when compared to well known procedures such as Maximum Likelihood or the Burg estimator.