June 2009

• Thursday, 25 June 2009, 1-2pm*
  Host: Dhananjay Thiruvady
  Title: Beam-ACO for TSP with Time Windows
  Abstract: The travelling salesman problem with time windows is a difficult optimization problem that appears, for example, in logistics. Among the possible objective functions we chose the optimization of the makespan. For solving this problem we propose a so-called Beam-ACO algorithm, which is a hybrid method that combines ant colony optimization with beam search. In general, Beam-ACO algorithms heavily rely on accurate and computationally inexpensive bounding information for differentiating between partial solutions. In this work we use stochastic sampling as an alternative to bounding information. Our results clearly demonstrate that the proposed algorithm is currently a state-of-the-art method for the tackled problem.


• Wednesday, 17 June 2009, 1-2pm*
  Host: Oliver Bown
  Title: Ecosystemic Approaches to Generative Music
  Abstract: Computational creativity investigates how computers can originate creative outputs. At CEMA, we are investigating ways of exploiting the creativity of natural evolution in artistic creative practices. In this area, the notion of interactive aesthetic selection has been around for a long time and has been used successfully in certain limited cases. Creative evolutionary software that exploits more diverse coevolutionary processes has had little impact in the arts for reasons of its complexity and the difficulty of appropriately coupling the evolutionary process in the software with processes leading to creative artistic outcomes. My research is concerned with studying theories of creativity in human and natural systems in order to understand ways in which nature-inspired systems can play a role in human creative processes, and therefore how software can incorporate evolutionary creativity into an individual's artistic creative workflow, or more widely in artistic social domains. I am currently focusing on tools to help design interesting multi-agent systems in specific aesthetic contexts. My special interest is in the area of generative music, and I will present work on 'sonic ecosystems': generative sonic artworks that situate an evolving population of agents in a sonic environment. I will discuss the theories of human, natural and computational creativity that form the background to this work, followed by my research into building sonic ecosystems, including two frameworks that I am developing on the way: one for multi-agent based creative computing, and one for computer music.


• Tuesday, 9 June 2009, 1-2pm*
  Host: Iris Yan (Faculty of Engineering)
  Title: A Traveling Time Prediction Based Vertical Handover Decision Algorithm between Cellular Networks and WLANs
  Abstract: The emerging IEEE 802.21 standard supports algorithms enabling seamless handovers between different network types, such as the 3G wide area cellular telephone networks, IEEE 802.11 WLANs and IEEE 802.16 WiMAX. In future wireless systems, the integration of various wireless network technologies provides the mobile users higher bandwidth at reduced cost. In order to enjoy such benefits, effective vertical handovers between two network technologies become a critical issue. In our proposal, a traveling time prediction based algorithm is developed to increase the efficiency of handovers between wide and local area networks and WLANs. The traveling distances inside the WLAN and the boundary area of the WLAN are estimated through the received signal strength measurements, and then compared against distance thresholds calculated from various network parameters. Handover decisions are based on the comparison results. The theoretical analysis and simulation results show that the proposed method enhances the performance of the handovers between wide and local area networks.


• Wednesday, 3 June 2009, 1-2pm*
  Host: Fabian Bohnert
  Title: Spatial Processes for Recommender Systems
  Abstract: Spatial processes are typically used to analyse and predict geographic data. In this talk, we give a brief introduction to spatial statistics, and develop a model based on Gaussian spatial processes for predicting a user's interests or item ratings within recommender systems. We present the theoretical framework for this model, and discuss efficient algorithms for parameter estimation. Our model was evaluated with a real-world dataset of time spans spent by museum visitors at exhibits (viewed as implicit ratings). In this application scenario, our model achieves a higher predictive accuracy than nearest-neighbour collaborative filters.

May 2009

• Monday, 25 May 2009, 1-2pm*
  Host: Daswin De Silva
  Title: Incremental Subspace Learning
  Abstract: The ability to generalise situations, problems and even dialogues has contributed immensely towards human development. Subspace learning and analysis encompasses research efforts to produce the same human generalisation process in machines using algorithms. A typical subspace algorithm attempts to project the original high dimensional feature space in a given problem or environment into a low dimensional subspace. This presentation will focus on an unsupervised learning algorithm for incremental subspace analysis. Incremental analysis sustains the pattern discovery process, as past patterns form the basis for the continuous learning process of the proposed algorithm. This is particularly gainful in application domains with a continuous dimension, such as time. Experiments conducted in feature recognition, a popular subspace learning domain, has further established the suitability of the proposed algorithm for continuous subspace learning.


• Wednesday, 20 May 2009, 1-2pm*
  Host: Marc Cheong
  Title: Decomposing Co-occurrence Matrices with Tchebichef Orthogonal Polynomials for Traditional Textile Motif Classification and Recognition
  Abstract: The Grey-Level Co-occurrence Matrix (GLCM) is a powerful yet simple method by Haralick et al. (1973) used in texture recognition and classification. 36 years down the line, it has been the basis for numerous research works on image processing, with applications from automated image retrieval to biomedical and satellite imaging. Mr Kar-Seng Loke and I have been working on a novel method of decomposing a GLCM using orthogonal polynomials and utilizing the resulting coefficients as a better descriptor for patterns compared to Haralick's original statistical analysis on GLCMs. We have adapted the GLCM method to handle color images and performed decomposition using Tchebichef discrete orthogonal polynomials (with the discrete cosine transform as a comparison); coupled our work with existing clustering techniques and applied it in recognizing and classifying traditional Malaysian 'batik' and 'songket' textile motifs. My talk will center around some background information on our research and our findings from the papers we have published on this subject; and I plan to make it more informative rather than technical in nature. And last but not least, feel free to sit back and enjoy some pizzas and a coffee :)


• Wednesday, 13 May 2009, 1-2pm*
  Host: Arun Mani
  Title: What are Matroids?
  Abstract: Recall from elementary linear algebra that a set of vectors are linearly dependent if they satisfy a simple linear equation among themselves. Matroids formalize the combinatorial properties of this linear dependence. These structures also turn out to be useful because if we can model our problem as a matroid, then we can use a simple greedy optimization algorithm to solve the problem. In this talk, I will give a brief tour of matroids and discuss (time permitting) how they are relevant to my research. I'll try to keep the math requirements to simple set theory and linear algebra (linear dependence and ranks).


• Wednesday, 6 May 2009, 1-2pm*
  Hosts: Christian Guttmann and Ian Thomas (Faculty of Medicine)
  Title: A Demonstration of the Agent-Based Intelligent Collaborative Care Management System
  Abstract: The aim of our research is to provide a unified model for the composition and management of consumer care services. We identify design, composition, distribution and management as key stages of this model and propose an Intelligent Collaborative Care Management (ICCM) System as its realisation. The distribution and management stages are implemented as multi-agent systems. Agents in the distribution stage carry out domain-specific negotiation and distribution processes for the assignment of tasks in the care plan. The agents in the management stage aim for failure prevention and adherence support in contrast to failure recovery in planning. The key to failure prevention is to identify what has to be carried out to prevent care plan failures. The healthcare domain is used to demonstrate the ICCM system.


• Friday, 1 May 2009, 1-2pm**
  Host: Ee Hui Lim (Faculty of Engineering)
  Title: Infinite Gaussian Mixture Modelling for Multi-Structure Segmentation
  Abstract: One of the solutions to terrestrial building recontructions is to extract planar data (from man-made structures) and then geometrically fit locally delimited planes to the data. Robust segmentation, especially the Random Sample Consensus (RANSAC) algorithm is most widely used for geometric modelling. However, there has not been a successful demonstration of segmentation of complicated outdoor data using only robust segmentation. Clustering provides an alternative approach to segmenting the data simultaneously. However, the clustering approach has not been very popular in multi-structure segmentation, due to the low tolerance in gross outliers and unknown actual number of planes. We proposed to remove non-planar data by classifying the raw outdoor terrestrial point clouds, making the data suitable for clustering with the Infinite Gaussian Mixture Model into different locally delimited planes. We have conducted preliminary experiments that have shown the applicability of the algorithm in robust multi-structure segmentation.

April 2009

• Tuesday, 21 April 2009, 1-2pm**
  Host: Bin Liu
  Title: A Comparative Study in Nonparametric Inference
  Abstract: Nonparametric inference is free of distributional assumptions and estimates probability distributions in an infinite-dimensional space. As with its parametric counterpart, nonparametric inference can be separated into two classes: frequentist and Bayesian approaches. Kernel density estimation (KDE) is an important method in nonparametric inference. We introduce some asymptotic statistics in nonparametric inference. We also study nine bandwidth selection schemes for kernel density estimation in Naive Bayesian classification.


• Friday, 17 April 2009, 1-2pm**
  Host: Mauro Bampo
  Title: Innovative Ideas in Computation
  Abstract: We will watch videos of two or three recent talks presenting new and fascinating ways to perform computation without the use of conventional computers. Discussion will follow.


• Wednesday, 8 April 2009, 1-2pm**
  Host: Amiza Amir
  Title: Content Recognition for Solving File Pollution Problems within P2P Networks
  Abstract: This talk will explore the use of content recognition as an effective method to unravel file pollution problems within P2P networks. File pollution is a major problem in a P2P file sharing system. Pollution in a P2P file sharing system is said to have occurred when a shared file or version either has a different content from what is expected; contains malicious object; or is corrupted and damaged due to transmission failure or bugs in software and machines which are used to create and manipulate it. The problem is more complicated considering that P2P networks work without a central authority or a server.


• Friday, 3 April 2009, 12:15-1:15pm** (as part of the FIT HDR workshop)
  Host: Jenny Kashmirian
  Title: A Simulation Model for Transient Hydrogen Bond Networks
  Abstract: At first glance people and plankton appear very different but at the molecular level they are remarkably similar. They use the same types of chemical molecules, similar principles for cellular organization and have a chemistry that depends on water. Studying the general principles of cellular organization not only gives an insight into the origin of life but to the organization of complex systems. My thesis aims to investigate one organizational principle at the molecular level, namely transient hydrogen bond networks. This task brings together the fields of artificial life, complexity and chemistry. From a computer science viewpoint this involves development of a model which captures the essence of the hydrogen bond network while abstracting out the detail. This could lead to a greater understanding of the properties and role of transient networks in biological systems. It could also contribute to understanding how network structure influences its behavior, a field of study in its infancy. This talk will cover the background of this research and show progress towards a working hydrogen bond model.

March 2009

• Tuesday, 31 March 2009, 1-2pm**
  Host: Anang Hudaya Muhamad Amin
  Title: Integrating Sensory Data within a Structural Analysis Grid
  Abstract: State-of-the-art and high investment structures such as aerospace vehicles, offshore installations, maritime vessels, and critical infrastructure require strenuous analysis, design, and monitoring processes to ensure their operability and safety. In doing so, vast amount of data needs to be analysed repeatedly where the processing steps follow well defined work-flows. The current practices for managing a structure’s lifecycle usually involve separate schemes for analysis, design, and monitoring and are thus not able to utilise the wealth of information created during the analysis and design phase towards monitoring and vice versa. Furthermore these practices do not include real time detection of approaching critical conditions. In this talk, we propose a proto-type design of an integrated grid-sensor network framework which can support end-to-end analysis, design, and monitoring work-flows for rapid structural engineering applications.


• Wednesday, 25 March 2009, 1-2pm**
  Host: Michael Wybrow
  Title: Scrolling Behaviour with Single- and Multi-Column Layout
  Abstract: The standard layout model used by web browsers is to lay text out in a vertical scroll using a single column. The horizontal-scroll layout model – in which text is laid out in columns whose height is set to that of the browser window and the viewer scrolls horizontally – seems well-suited to multi-column layout on electronic devices. We describe a study that examines how people read and, in particular, the strategies they use for scrolling with these two models when reading large textual documents on a standard computer monitor. We compare usability of the models and evaluate both user preferences and the effect of the model on performance. Also interesting is the description of the browser and its user interface which we used for the study.


• Wednesday, 18 March 2009, 1-2pm**
  Host: Christopher Chua
  Title: Pheromones for Swarm Robotics
  Abstract: Social insects such as ants co-ordinate their actions in a decentralised manner by marking their environment with chemicals called pheromones. This type of co-ordination brings with it simplicity and robustness, but is difficult to translate into the mechatronical world. A prototype sensor mounted on a small robot will be described and demonstrated. Issues such as pheromone material selection, sensor design trade-offs and dispersion of the pheromones will also be discussed.


• Wednesday, 11 March 2009, 1-2pm**
  Host: Simon Thompson (British Telecommunications Research)
  Title: British Telecommunications (BT) Research – Organisation and Orientation of a European Telco's Research Department
  Abstract: In this presentation I will describe what the current role of BT Research is, how we are organised and funded and how we deliver to our business. I will describe the process of devising, proposing and executing research projects in our company.


• Wednesday, 4 March 2009, 1-2pm**
  Host: Yee Ling Boo
  Title: Mining Multi-Modal Crime Patterns at Different Levels of Granularity using Hierarchical Clustering
  Abstract: The appearance of patterns could be found in different modalities of a domain, where the different modalities refer to the data sources that constitute different aspects of a domain. Particularly, the domain of our discussion refers to crime and the different modalities refer to the different data sources such as offender data, weapon data, etc. in the crime domain. In addition, patterns also exist at different levels of granularity for each modality. In order to have a thorough understanding of a domain, it is important to reveal the hidden patterns through the data explorations at different levels of granularity and for each modality. Therefore, this talk will present a conceptual model to identify patterns that exist at different levels of granularity for different modes of crime data. A hierarchical clustering approach – Growing Self-Organising Maps (GSOM) – has been deployed. Furthermore, the conceptual model is enhanced with experiments that exhibit the importance of exploring data at different granularities.

February 2009

• Wednesday, 25 February 2009, 1-2pm**
  Host: Jens Kötters
  Title: Context Graphs
  Abstract: Suppose that you have a number of objects (e.g., animals, events, webpages, ...) and some attributes in terms of which the objects can be described. In this talk, I introduce a graph model – termed "context graph" – in which the objects (being the nodes of the graph) are connected on the basis of similar descriptions. Objects with common attributes (e.g., all travel guides under $20, if the objects are books) are located in the same part of the graph. Intuitively, the graph can be seen as a kind of semantic map. So far I have considered this model mainly for Information Retrieval (IR). The main advantage of this model for IR is that it supports the combination of querying and navigation. Social network analysis is probably another application.


• Wednesday, 18 February 2009, 1-2pm**
  Host: Subrata Chakraborty
  Title: Selecting the Most Suitable MADM Method for Different Problem Settings
  Abstract: Multiattribute decision making (MADM) methods are widely used in ranking and selection problems, where several alternatives need to be ranked with respect to several decision criteria. With the development of many MADM methods, selecting the most suitable one for any particular decision problem has itself become an MADM problem. In this talk the basic aspects of an MADM problem will be introduced along with various challenging issues in solving an MADM problem. Some experimental results will also be presented which will provide some valuable insights for selecting the most suitable method under certain decision settings.


• Wednesday, 11 February 2009, 1-2pm**
  Host: Benjamin Porter
  Title: Organic Form Synthesis through Morphogenetic Simulation
  Abstract: Modelling the geometry of organic forms using traditional CAD or animation tools is often difficult and tedious. Different models of morphogenesis have been successfully applied to this problem; however many kinds of organic shape still pose difficulty. This talk introduces a novel system, the Simplicial Developmental System (SDS), which simulates morphogenetic and physical processes in order to generate specific organic forms. SDS models a system of cells as a dynamic simplicial complex in two or three dimensions that is governed by physical rules. Through growth, division, death, and movement, the cells transform the geometric and physical representations of the form. The actions of the cells are governed by conditional rules and communication between cells is supported with a continuous morphogen model. Results are presented in which simple organic forms are grown using a model inspired by limb bud development in chick embryos. These results are discussed in the context of using SDS as a creative system.


• Wednesday, 4 February 2009, 1-2pm**
  Host: Toby Smith
  Title: Incremental Learning with Self-Organising Maps
  Abstract: Incremental clustering algorithms are clustering algorithms which as well as having the ability to map data clusters as they appear in a data stream, they also maintain something of a memory of past clusters (Allowing new clusters to be compared with historical clusters). One of the primary uses for incremental clustering algorithms are for longitudinal data analysis. This talk will provide a brief synopsis of a couple algorithms based on a type of neural network called a Self-Organizing Map (SOM), and present a new algorithm for incremental learning algorithm for longitudinal data analysis.

January 2009

• Wednesday, 28 January 2009, 1-2pm**
  Host: Christopher Mears
  Title: Exploiting Symmetry in Constraint Programming
  Abstract: Symmetries are present in many difficult combinatorial problems. In constraint programming – a technique used for solving such problems – the presence of symmetries can cause much wasted time and effort. I will give a brief overview of constraint programming and symmetries, and describe some ways of improving the resolution of symmetric problems.


• Wednesday, 21 January 2009, 1-2pm**
  Host: Dhananjay Thiruvady
  Title: Incorporating Constraint Programming in a Hybrid Beam-ACO Algorithm
  Abstract: We have previously shown that constraint programming (CP) integrated with ant colony optimization (ACO), CPACO, is effective on problems with hard constraints. However, the run-time overhead of this algorithm is prohibitive. By considering Beam search we are able to avoid repeated constraint propagation needed by CPACO and hence conduct improved searches with efficient run-times. The algorithms are tested on a single machine job scheduling problem.


• Thursday, 15 January 2009, 1-2pm**
  Host: Minh Duc Cao
  Title: Biological Compression and Its Applications
  Abstract: In this talk, I am presenting the eXpert Model (XM), a novel algorithm for compression of biological sequences. Not only the XM outperforms most existing compression algorithms on the standard datasets, it has shown the ability to discovery patterns in sequences of biased nucleotide composition where spurious patterns are often observed because the bias leads to coincidental matches. If time permits, I will demonstrate how the XM can be used for performing various bioinformatics tasks including sequence alignment and phylogenetics.