Waikato University crest

Department of
Computer Science
Tari Rorohiko

Computing and Mathematical Sciences

Seminars Presented in 2005

Events Index

Teaching international students

Dorothy Spiller
TLDU, The University of Waikato
Tuesday, 13th December 2005

The last seminar that Dorothy gave here (on teaching issues) started so many heated discussions, that the participants decided that it would be interesting to focus on a specific aspect in a second seminar.

Dorothy will discuss ways to handle classes with large proportions of international students. It is definitely going to be interesting!


A Fuzzy Ontology for medical information retrieval

Dave Parry
AUT, Auckland
Tuesday, 22nd November 2005
A mapping between query terms and members of an ontology is usually a key part of any ontology enhanced searching tool. However many terms used in queries may be overloaded in terms of the ontology, which limits the potential use of automatic query expansion and refinement. In particular this problem affects information systems where different users are likely to expect different meanings for the same term. This paper describes the use of a "Fuzzy Ontology" which uses fuzzy relations between components of the ontology in order to preserve a common structure. The concept is presented in the medical domain, using a modification of the Medical Subject Heading (MeSH) hierarchy. By adding a value for degree of membership to each relation for each user or group of users then the recovered documents from ontology mediated search can reflect the likely information need. This seminar will discuss the fuzzification scheme and its representation using XML.

This work is part of Dave's PhD study on "An Intelligent system for finding useful Medical information", which he hopes to submit very soon!


How to successfully apply for research funding

Mark Apperley
SCMS, University of Waikato
Tuesday, 15th November 2005


Promoting student learning through assessment and classroom participation

Dorothy Spiller
TLDU, University of Waikato
Tuesday, 11th October 2005
Dorothy will give a brief initial talk about topics concerning tertiary teaching. We will then have the opportunity for an extensive question and answer session.


Enrollment analysis using data mining

Rahul Tikekar
Southern Oregon University, USA
Tuesday, 4th October 2005
The number of computer science majors at my university (as with other universities in the US) has dropped dramatically in the last four years. The overall university enrollment has not changed much. This means that the students are seeking other majors.

In this talk I will present a preliminary study, done as part of a master's project, to understand where the students are going and why.

If time permits I will also present a hodgepodge of other topics in which I am currently interested.


Peer assessment and beyond

John Hamer
Auckland University
Tuesday, 27th September 2005
Recently, a number of groups at the University of Auckland have adopted peer assessment as a routine activity in undergraduate courses. This has been supported by a web-based application, Aropa, that greatly simplifies most of the required administration tasks. In this talk, I will describe what the software does and reflect on our experiences to date. This is a non-technical talk. Teaching staff from departments outside Computer Science are particularly welcome.


Producing good software fast

Ewan Tempero
University of Auckland, New Zealand
Thursday, 11th August 2005
Figuring out how to produce good software fast could be said to the the goal of software engineering research, but to meet the goal several questions must be answered, including: what does "good" mean? how do we measure "goodness"? if we determine the software is not "good", what do we do to improve it?

In this talk, I will give an overview of my research in the context of this goal. I will discuss the aspects of software quality I am interested, namely maintainability, testability, and reusability, and the issues involved in measuring such aspects, and I will present some results on measuring and improving software quality, including a variant on the old notion of software coupling.


Combining model-based and instance-based learning for first order regression

Kurt Driessens
University of Leuven, Belgium
Tuesday, 2nd August 2005
The introduction of relational reinforcement learning gave rise to the development of several first order regression algorithms. So far, these algorithms have employed either a model-based approach or an instance-based approach. As a consequence, they suffer from the typical drawbacks of model-based learning such as coarse function approximation or those of lazy learning such as high computational intensity.

In this talk I will introduce a new relational regression algorithm that combines the strong points of both approaches and tries to avoid the normally inherent draw-backs. By combining model-based and instance-based learning, an incremental first order regression algorithm is produced that is both computationally efficient and results in better predictions earlier in the learning experiment.

The talk is a trial run for a conference talk and will therefor be only 20 minutes long.


Graphical abstract help

Jeff Huang
University of Illinois, Urbana-Champaign
Wednesday, 6th July 2005
We explore the use of abstracted versions of screenshots as part of an interface to support giving help. Graphstract, the software implementation of this graphical help system, extends the ideas of textually oriented Minimal Manuals to the use of screenshots, enabling multiple small graphical elements to be shown in a small space. This enables the user to get an overview of a complex sequential task as a whole. Graphical hints, such as jagged edges, red dots, and icons are also explored. The idea has been developed by iterative prototyping. In cases where the minimalist help is insufficient, ways of providing more detailed information on demand are investigated.


Interactive visualization and navigation using the hyperbolic space

Jörg Walter
Technische Fakultät, Universität Bielefeld, Germany
Tuesday, 14th June 2005
We present the Hybrid Hyperbolic Data Viewer (HHDV) which combines methods for the interactive visual data mining of large collections of data. Both, Hyperbolic Multi-Dimensional Scaling (HMDS) and Hyperbolic Self-Organizing Maps (HSOM) employ the extraordinary advantages of the hyperbolic plane (H2): (i) the underlying space grows exponentially with its radius around each point - ideal for embedding high-dimensional (or hierarchical) data; (ii) the Poincare model of the H2 exhibits a fish-eye perspective with a focus area and a context preserving surrounding; (iii) the mouse binding of focus-transfer allows intuitive interactive navigation.

The HMDS approach extends multi-dimensional scaling and generates a spatial embedding of the data representing their dissimilarity structure as faithfully as possible. It is very suitable for interactive browsing of data object collections, but calls for batch precomputation for larger collection sizes.

The HSOM is an extension of Kohonen's Self-Organizing Map and generates a partitioning of the data collection assigned to an H2 tessellating grid. While the algorithm's complexity is linear in the collection size, the data browsing is rigidly bound to the underlying grid.

By integrating the two approaches we gain the synergetic effect of adding advantages of both. And the hybrid architecture uses consistently the H2 visualization and navigation concept.

An application to a text mining example involving the Reuters-21578 text corpus is presented besides several other applications in the domain of text, music, and images.


Learning from huge data sets by SVMs

Vojislav Kecman
Auckland University
Tuesday, 7th June 2005
The seminar introduces the basics of a 'novel' learning approach known as the support vector machines (SVMs) and it discusses a novel iterative approach in solving (QP based) SVMs learning problem when faced with huge data sets (say several hundreds thousands, or more, training data pairs, i.e., measurements i.e., samples). It will be shown that the SVMs learning is related to both the classic Kernel AdaTron method and Gauss-Seidel iterative procedure for solving a system of linear equations with constraints. Comparisons with an SMO based algorithms will be given. Expected audience are all interested in machine learning, as well as the users and developers of the SVMs learning tools.


The 'WorldTree' Project: a case study in 'postmodern software design'

Colin Beardon
The University of Waikato
Tuesday, 31st May 2005
The seminar will open with a showing of the 'WorldTree' digital art work that was a finalist in the recent Vodaphone Digital Art Awards (4 minutes). I will describe the methods I use to develop works of this kind, referring both to traditional creative practices and development within a digital environment. When developing in a non-digital context there are many useful practices (sketching, collage, abstraction, etc.) but these have no obvious counterparts within software development. I will discuss some of the issues raised when trying to talk about my digital work and try to indicate what I mean by 'postmodern software design' - i.e. the design of software that does not (yet) have a meaning.


The virtuous cycle: agile programming in a small team

John Cleary
ReelTwo Ltd and The University of Waikato
Tuesday, 24th May 2005
A look at agile programming practices in a small team of 6 programmers doing software product development. The talk will look at those practices successfully used within the company, including; shared code, continual compile and build, and unit testing. It will also look at ways of ensuring that a system can be easily tested and at parts of the programming process which are difficult to automate using agile programming.


Possible research cooperation with Fonterra?

Margaret Jefferies and Tony Smith
University of Waikato
Tuesday, 17th May 2005
Fonterra is interested in sponsoring research in cooperation with the department. Last week, Tony and Margaret went to a meeting with the Fonterra people. They will tell us about the outcome of the meeting and what the plans are to make this research cooperation happening.

Do all come along - we are talking real research money here.


Multimodal dialogue systems

Arne Jönsson
Linköping University, Sweden
Thursday, 10th May 2005
The talk will present the research on multimodal dialogue systems we are doing in Linköping. I will use dialogue systems that we have developed at our research lab to illustrate some issues in multimodal interaction, including ontologies in dialogue systems, how information extraction can be used, user modeling, new modalities and design patterns for development of dialogue systems.


Is software engineering really engineering?

John Grundy
University of Auckland, New Zealand
Tuesday, 3rd May 2005
While software engineering methods, tools and processes continue to mature and become more widely adopted for software projects world-wide, there is still debate on how much "Engineering" is embodied in the discipline. This talk looks not so much at what is similar between Software Engineering and other, both traditional and non-traditional, branches of Engineering, but what is inherently different in the realm of software construction and therefore on the application of engineering approaches to software. This talk focuses on practical issues and examples, but also reflects on theoretical concepts and future directions of Software Engineering, but also the place of Engineering in general within 21st century business and society.


Multi-device user interfaces and architectures for information systems

John Grundy
Technical University of Munich, Germany
Tuesday, 5th April 2005

Multi-device user interfaces (or MUIs) are intended to run on a variety of disparate display devices e.g. desktop PC, laptop, wireless PDA, mobile phone, large E-whiteboard etc. In this talk I will provide a motivation for MUIs and give some examples of MUI architectures and prototype information systems that we have developed. These include systems for the travel industry, health informatics, and on-line retail systems.

I will try and identify some of the key challenges in engineering MUIs and outline some of the approaches we have taken to addressing some of these challenges.


Machine learning and data mining in bioinformatics: a personal view

Stefan Kramer
Technical University of Munich, Germany
Thursday, 31st March 2005

In the talk I will give a survey of machine learning tasks in bioinformatics, techniques used to address them and our own work in this area. Machine learning applications in bioinformatics are characterized by (1) high-dimensional and structured data, (2) the need for integrating diverse data from heterogeneous sources and data mining techniques, (3) structured output (not just binary classification), and (4) data quality issues. Sample applications and solutions will be presented to illustrate these points.


Supporting informal collaboration in groupware

Carl Gutwin
University of Saskatchewan, Canada
Tuesday, 22nd March 2005

Shared-workspace groupware has not become common in the workplace, despite many positive results from the research lab. In this short talk I will look at one reason for this lack of success: that most shared workspace systems are designed around the idea of planned, formal collaboration sessions � when much of the collaboration that occurs in a co-located work group is informal and opportunistic. To support informal collaboration, groupware must be designed and built differently. I will introduce the idea of communityware, in which groupware is organized around groups of people working independently, rather than shared applications, documents, or virtual places.

Communityware provides support for three things that are fundamental to informal collaboration: awareness of others and their individual work, lightweight means for initiating interactions, and the ability to move into closely-coupled collaboration when necessary.


Expressive event filtering in distributed systems - non-canonical filtering

Sven Bittner
University of Waikato
Tuesday, 22nd March 2005

Event Notification Services (ENSs) offer a push-based approach to access information. Thus, in contrast to traditional pull-based information systems they allow for continuous information demands: Interests are specified once with the help of continuous queries which are executed whenever new information arrives at the system. This information is sent to the ENSs by various publishers in form of event messages. Applications for ENSs can be found in digital libraries, e-commerce, facility management and general sensor networks.

Current matching approaches for high performance ENSs mostly allow for the formulation of very restricted continuous queries. Hence, arbitrary Boolean filter expressions have to be transformed into canonical expressions, e.g., disjunctive normal forms, and need to be treated as several conjunctive queries. This technique is known from database systems and allows us to apply more efficient filtering algorithms. Since ENSs are the contrary to traditional database systems, it is questionable if filtering several canonical queries is the most efficient and scalable way of dealing with arbitrary expressions.

In this talk, I will present current solutions for an efficient matching in ENSs and identify their benefits and drawbacks. I also introduce a novel non-canonical matching approach and show the results of a comparison of this approach to a well-known canonical solution.


Supervisory control and automatic model generation for coordination of industrial robots

Hugo Flordal
Chalmers University of Technology, Gothenburg, Sweden
Tuesday, 8th March 2005

The Supervisory Control Theory is a theory developed for verification and synthesis of processes and supervisors in the field of automation - "formal methods for manufacturing systems". Programming of Programmable Logic Controllers (PLC:s) in industry is a bottleneck in the development of new production lines. Traditionally, the programming is performed manually, making it a tedious, error-prone task. The PLC-code being man-made also greatly complicates the maintenance and adjustments of such production lines. To allow for faster development of new and old production lines and for mass customisation of products, a greater flexibility is needed in this respect.

A method for automatic generation of models for coordination of industrial robot cells is presented. The setting is a cell with a number of robots working closely, with a risk of collisions and a need for coordination. Each robot's task is specified as a set of targets that the robot should visit in arbitraty order - a scenario quite like for example spot welding in car industry. Finite automaton models of allocation and release of critical spatial volumes that the robots share, as well as models of the robots' possible movements are automatically extracted from a 3D simulation environment. This includes explicitly calculating the intersection between the robots' work envelopes - the spatial volumes where collisions may occur - and simulating the robots' collisions with these. The number of possible sequences of operation that could be considered for a robot is obviously factorial in the number of states, but can be efficiently represented as a set of automata using a number of states polynomial in the number of targets.

The models can be used either to generate a supervisor using the SCT framework, allowing as much freedom as possible to the robots thus "maximising flexibility", or a controller using optimization based on the simulated times for the robots' motion, "minimising work cycle time". Based on either of these results, PLC-code can be generated that guarantees collision and blocking avoidance in the robot cell.


Robot localization using appearance

Matjaz Jogan
University of Ljubljana
Monday, 20th February 2005

Appearance-based visual learning and recognition techniques that are based on models derived from a training set of 2D images are being widely used in computer vision applications. In robotics, they have received most attention in visual navigation and serving. In this talk I will present a framework for visual self-localization of mobile robots using a parametric model built from panoramic snapshots of the environment. In particular, I will focus on solutions to lifelong learning and to the problems related to robustness against occlusion, illumination conditions and invariance to rotation of the sensor. I will stress the advantages and the drawbacks of appearance-based representations in terms of a cognitive vision system capable of robust operation and lifelong learning.


Introducing the M2K framework for music IR research and evaluation

Stephen Downie
University of Illinois at Urbana-Champaign
Tuesday, 18th January 2005

The IMIRSEL (International Music Information Retrieval Systems Evaluation Laboratory) project being undertaken at the University of Illinois at Urbana-Champaign (UIUC) provides an unprecedented platform for evaluating Music Information Retrieval (MIR) and Music Digital Library (MDL) techniques. The standardized research collection being deployed represents a large and diverse corpus of musical examples, which we are hosting in our secure environment for use in evaluating MIR/MDL algorithms. Grid services and the National Center for Supercomputing Applications's (NCSA) "D2K" machine learning environment provide a powerful, high-performance, and secure framework for designing, optimising, and executing complex MIR/MDL evaluation applications. The "M2K", or "Music-to-Knowledge" suite of tools being developed extend the D2K enviroment to allow MIR/MDL researchers the ability to rapidly prototype and evaluate their MIR/MDL systems and algorithms. IMIRSEL, and its Human Use of Information Retrieval Systems (HUMIRS) sub-project, will deliver a set of community resources for researchers who would otherwise not be able to afford the content rights and computational resources to carry out large-scale MIR/MDL evaluations.


Metadata handling in collaborative environments

Maria Oelinger
University of Duisburg-Essen
Friday, 7 January 2005
Collaborative learning in intelligent educational environments can be supported using rich metadata. The modelling and usage of this metadata may be implemented using a centralised archive of semantically described learning objects. The objective is to support the learners by fostering their research skills and making (re)use of existing material as a base for enhanced learning. The metadata can reflect the scientific background of a user or the research context of a learning project. Maria will present selected results of the research project Coldex for computer-supported collaborative learning at the University of Duisburg-Essen, Germany.


Events Index