An introduction to the life of a dot-com in the post-dot-com area. Starting first with the background in dvd-rentals in the global market and its conception and growth in the UK and Nordic markets. 5 years ago there were about 15 small dvd-rental companies in the UK, now there are only 3 - with LOVEFiLM representing the easy majority of online dvd-rental (along with Amazon and Blockbuster). Will then discuss the implications of an international multi-lingual company. This will focus on technologies, mergers, take-overs and then moving onto my personal experiences working with the company. And then (if time permits) close with the future of the media markets in general and how downloading is changing the face of the home cinema market.

In data mining there is a kind of dataset with binary response variable and numerical predicator variables. Logistic regression is a common and powerful method for dealing with this kind of dataset. However, when the datasets have the properties of collinearity, nonlinearity or interactions, the decision boundary created by logistic regression becomes inaccurate. The common method for solving this problem is to recursively partition the data space so that each subspace can be solved well by a single logistic regression. The latest Logistic regression tree model is called LOTUS. This method can build a tree model very fast. However, it sometimes builds a tree which is just a little better than using logistic regression alone. The aim of my project is to design and build a logistic regression tree model which can make a good improvement in accuracy for most cases and build the tree in a feasible time.

Do you see the world in 3D or just as a sequence of pictures? Does it matter? We start with a couple of 'simple' problems in visualisation and go on to show how truly 3D thinking can make some tricky maths much simpler.

This is a much cut down version of the tutorial I did in SIGGRAPH in 2001. I have done it before at a Graphics meeting but not since 2002.

A short business meeting at which we work out a plan how to divide and share certain tasks now that Sui-Ling has gone.

Research into the human circulatory system suggests that the shape of red blood cells can be used to diagnose or even predict certain illnesses, such as ME and MS. Right now, haematologists have to spend long hours counting each cell by hand. The aim of the project I'm involved in is to automate this process.

In my talk I will present my approach to the project's first step: segmenting the SEM images into individual cells. I use a guided contour tracing approach, which I will explain and illustrate with a short demonstration. I'll also present some preliminary results. Afterwards, there will of course be ample time for any and all burning questions you might have, or, failing that, afternoon tea.

My PhD project is on animating a dog using physical (newtonian) models. Existing models are often based on robotics rather than on biophysical models. This talk will be about what difference we are trying to make. As the project is currently in the stage of developing a physics simulation environment, the talk will contain a brief overview of what rigid body simulation is about, and in more detail a method of simulating linkages (chains) of rigid bodies will be described.

A short talk about attempting to produce a program to play a credible game of Scrabble.

Right Hemisphere are the creators of well-known software such as Deep Paint and Deep Creator and market themselves as the "leading provider of Product Graphics Management (PGM) software". They are based in Auckland, Silicon Valley, and have recently established their European headquarters in Munich, Germany.

Jed will give a general overview of Right Hemisphere. He will also give an introduction to the world of Product Graphics Management and CAD visualization, the 3D and CAD graphics industry on the whole and the skills and opportunities available to those interested, be it for CAD, Games, 3D, Visualization, etc. Questions are welcome throughout the talk.

Website: http://www.righthemisphere.com/

In artificial intelligence, an evolutionary algorithm (EA), as a subset of evolutionary computation, is a popular technique of metaheuristic optimization. During the last two decades, simulating and evolving of virtual organisms has been the main target of several different projects, from which one of the best known is Karl Sims's work "Evolving Virtual Creatures", published in 1994.

As a part of my master degree studies at Charles University in Prague, Czech Republic, I have been a member of the group working on the project named Evolution of Robotic Organisms (ERO), which two main goals are:
1. Create an open source 'reimplementation' of Karl Sims's work (extending it slightly) using today's computers, and provide the result to the public.
2. Develop a distributed computation environment that would allow researchers easily create and execute distributed computations, especially evolutionary computations, giving the opportunity to build it upon existing EA scheme.

I will present the approach of our group, discuss how we were inspired by Karl Sims's work, and show you some fundamentals of the ERO project.

Osteoporosis degrades the bone density which can lead to fracture (or deformity). This pilot study applies Digitized Video Fluoroscopy to study the motion of the spine of osteoporotic fracture and looks into how these fractures affect the motions. This will provide some understanding on the biomechanics of the fractured spine.

The "core gamer" market is saturated. If the games industry is to continue along the growth curve experienced over the last decade, then there need to be serious efforts made to reevaluate the market and our delivery of game content.

A key issue here is breaking the stereotype of the "gamer" by providing experiences that address the needs of the wider market. Nintendo are making a bold move to make games more accessible through their new motion sensor technology. The "Casual Games" market is exploding online through PC and XBOX marketplaces. Singstar is now a household name. Let's take a moment to evaluate how new technology and new thinking can not only enhance the experience of the core gamer, but in fact help to find and engage new players.

** PLUS **
The premiere screening of the long-awaited stop motion animation by Alexis Angelidis and Damon Smith
Title: Omnivorous Insanity (Near ~Final Cut)
Synopsis: On their way to holidays, Tim & Tom are forced to take a day off at the beach. They will soon find out that food does not always come on a plate.
Website

We have recently been investigating what information a human processes when recognising reach/grasp hand actions. Various elements can be identified, hand trajectory, hand shape, body motion, but are any more important than others? I'll present some of the data from the pilot study conducted on people around the Graphics Lab and the preliminary results from our external subjects.

No, It's not computer vision, it's computer hearing! Following a brief overview of the system I'll delve into topics such as context sensitive pitch, note separation by pitch, and vibrato analysis.

What is good noise? To find out, we will take a look at the requirements for a noise function which is to be employed in graphics, and add a few extra criteria which should improve the quality of the noise. A method which computes an image of noise fulfilling these criteria, will then be described, and we will compare the result to what other well-established noise functions achieve. Some of the methods Geoff and I have tried out to achieve fast computation of good noise will be shown and, finally, I will propose a way to compute good noise on a GPU.

I'll present three methods to pose and animate characters based on an underlying skeleton hierarchy. The first method is commonly used in CG. The two others are generalizations: one prevents the surface from self-intersecting, and the other preserves the volume of the character. These properties also solve the candy-wrapper and collapsing-elbow effects. Because the methods are compatible, they can be combined simultaneously without additional computational cost for blending. Also, because they are of increasing complexity, they are suitable for a level of detail framework.

This work was postdoctoral research carried out at DGP with Karan Singh.

The 'Tartini' project aims to use the computer as a practical tool for singers and instrumentalists. Sound played into the system, a pc with a microphone attached to it running the Tartini software, is analyzed fast enough to create useful feedback for teaching or, at a higher level, for practicing musicians to refine their technique. Central to this analysis is the accurate determination of musical pitch. Another aspect is the analysis of notes that are being played with vibrato, a fast modulation in pitch that is typically used to add expression and vocal-like qualities to instrumental notes.

My assignments consists of automatically analyzing vibrato notes in Tartini, if possible in real-time, and give useful, real-time feedback to the musician who is playing the notes. I will talk about several approaches to vibrato analysis and the way I implemented them. There will also be a demonstration of the software!

Blood is a life sustaining fluid which circulates through our blood vessels, and caries oxygen and nutrients to the tissues and waste products away, to be removed by lungs, liver and kidneys. During illness abnormalities can develop in the cells of the blood. Different illnesses target different type of blood cells. With the help of technology the amount of blood cells in a sample of blood can be automatically counted, and because of the way of counting the amount of blood cells the size can be estimated as well. But not a lot can be said about the shape of the cells.

My assignment consists of analysing images from Red Blood Cells. There images are created with a Scanning Electron Microscope (SEM). First of all edge detection is performed on these images. These edges are used to generate ellipses that fit these edges as close as possible. Doing some extra calculation with the edges and ellipses I can reject bad ones. This should result in a number of good ellipses that fit Red Blood Cells closely. Now that indivdual Red Blood Cells can be selected, these can be used to extract 3D information about the shape of the cell. This will be done with Shape-from-Shading, mainly because SEM images have some nice properties that make it easier for Shape-from-Shading to come up with good results

Biomechanics studies the mechanics of living bodies, especially the forces exerted by muscles and gravity on the skeletal structure. The biomechanics group over Zoology studies animal locomotion. The research interest is passive dynamics in animal gaits. The approach is to build body structures to simulate the real movement in Matlab.

My project was to develop a GUI to cover up the coding scheme in Matlab, in which users can manipulate the build-in graphical representations of the body parts by mouse to construct body structures like what one could do with modelling systems like MAYA. The interface also provides viewing, saving, loading etc. system functions for the convenience of the users. If the dynamics is correctly set up, a click on the simulation button will show the movement of the body structure in the modelling window. The interface is to assist biomechanical analysis for non-mathematicians.

System speed complexity and the move to full or semi-custom silicon resulted the death of the garage hardware startup in the early '90s - things are about to change - a brief talk about the coming revolution and a short tutorial on how to do it yourself

Paul Crane "Multi-agent hierarchical system with learning"
Traditional AI for computer games leaves a lot to be desired. One of the things that needs work is to keep the game challenging. This project is about teaching the computer tactics so that the game remains interesting for the player.

Jeremy Weir "3D Character Animation"
I'll be talking about the steps I've taken to model, rig and animate a 3D character, using Maya.

What it's about: Logistic regression is a predictive modelling technique that tells you the odds of a particular outcome given a vector of features. It is used---quite heavily---in medical diagnosis, social sciences, and process control. I'll show you what it does, how it works, and some of its important limitations.

I'll also demonstrate an interesting connection between logistic regression and feed-forward neural networks. Come along if you think it might be interesting to see feed-forward implemented in two (2) lines of code.

We introduce a method for the animation of elastic and incompressible material. Our model is composed of two main parts: a motion equation for the vorticity at the interface of two continua, and an approximated Biot-Savart law to recover the velocity field. We apply this model to the animation of muscles for character animation, which relates our method to physically-based deformers.

Our technique requires the artist to place and size the muscles' geometry, and can produce a quick visualization of the dynamics with no heavy precomputation. We use an easy to implement explicit Euler solver, fast and stable for arbitrary long time steps.

"Critters" - Jeremy Weir
"Travelling Salesman" - Paul Crane

** Congratulations to Joe who is getting married today! **

Geoff will speak a little about his "bridge-building" successes;
Phil will talk briefly about High Dynamic Range photography using a normal digital camera;
Arthur and Phil will introduce the subject of Cross Platform Shader Languages: "You know the theory. But how do you actually code it?"

The talk will be about recent changes in the project, most notably a complete overhaul of the networking layer and the addition of a new interactive demonstration application.
After the talk, there will be an opportunity to try it out for yourself :-)

In this talk I will give a brief tutorial on the Mahalanobis distance followed by a brief look at how we can apply a similar distance to categorical variables. The Mahalanobis distance is a way of measuring distance in a non-Euclidean space where the variables are continuous, correlated and have different spreads.

It has an enormous number of applications in various classification contexts (eg face detection or face recognition, data mining etc) and is closely related to the technique of principal components. Generalising it to the case of categorical variables has proved problematic - until now (maybe)!

We introduce a novel method for describing and controlling a 3D smoke simulation. Using harmonic analysis and principal component analysis, we define an underlying description of the fluid flow that is compact and meaningful to non-expert users. The motion of the smoke can be modified with high level tools, such as animated current curves, attractors and tornadoes.

Our simulation is controllable, interactive and stable for arbitrarily long periods of time. The simulation's computational cost increases linearly in the number of motion samples and smoke particles. Our adaptive smoke particle representation conveniently incorporates the surface-like characteristics of real smoke.

With the academic collaboration of: Fabrice Neyret (INRIA, Grenoble); Karan Singh, Derek Nowrouzezahrai (DGP, Toronto)

The last year or so has been a busy one for Animation Research. We've developed new real time software for coverage of Gliding, Air Races, Air Traffic Control as well as expanding our ability to cover sports such as Golf, Cricket and America's Cup yachting. We've also created animations of the workings of the human body and disasters that level cities. On top of that we're having to grapple with all the problems that arise as a small company begins to grow. This will be a fairly informal presentation with more of an emphasis on question and answer, allowing us to concentrate more on what interests people!

Athol has been a long-standing member of the Dunedin Astronomical Society. He will present a talk on amateur astronomy and how the computer has changed the face of astronomy from space maps to space telescopes. The talk will include a little bit of history, observational techniques, basic equipment and some useful software. Ask questions at any time!

Detecting association rules with low support but high confidence is a difficult data mining problem. To find such rules using approaches like the Apriori algorithm, minimum support must be set very low, which results in a large amount of redundant rules. We are interested in sporadic rules; i.e. those that fall below a maximum support level but above the level of support expected from random coincidence.

In this paper we introduce an algorithm called MIISR to find a particular type of sporadic rule efficiently: where the support of the antecedent as a whole falls below maximum support, but where items may have quite high support individually. Our proposed method uses item constraints and coincidence pruning to discover these rules in reasonable time.

Animals are mechanically complex and have hugely complex neuromuscular systems that apparently help them to move rapidly and efficiently in complex environments. Despite this complexity, agile movement may rely on rather simple principles of biomechanical design and neural computation. In short, mechanical parameters of the body can be tuned to capture and recycle energy across phases of a movement, with the brain trimming these properties in real time. The required mathematical analysis is not entirely trivial, but computer graphics and animations make it possible to develop, explore and test numerical models interactively in a virtual world.

Physics-based simulation combined with 3D animation provides visualizations that can be appreciated by anyone, coupled to rigorous technical specifications (i.e. MATLAB code) that can be scrutinized by experts. I will show animations and live demonstrations of agility in virtual environments. I hope to convince you not only that there is some merit to the proposed principles of agility, but that agile legged robots - real and virtual - might be just around the corner.

The goal is to extend the current WW with the ability to recognize the hands of the user in "real-time". With this information the system can be enhanced in a way that the user can interact with the virtual objects.
The goal is achieved by using all three cameras to estimate the position of the complete upper body pose and derive the hand position from this body pose. The upper body pose is estimated with the use of a particle filter which "lives" in the 3D world. The particles are projected on each camera and each camera evaluates the particles individually and the particle filter uses the combined evaluations of the cameras to give each particle a weight.

Games, puzzles and a little bit of history...

Report on the outcome of this summer's gaming paper. Hopefully, we'll see a demo of a couple of the completed games.

The Moving Picture is a system which enables people to interact with a virtual character on a computer screen. People are detected in the live feed of a camera via a particle system; this data is then used to have the character react suitably. The presentation will cover the whole pipeline from visual detection to the behavior system of the character.

Introduction to the methods of image acqusition used in vidl2.

Some of you will recall Matthew breaking his collarbone just before Christmas during a stunt for his home movie. Well this is the movie...

In "Agent Arthur's Jungle Journey" a young agent is called upon to stop an evil distributor of a mind control potion. When the potion is lost and falls into enemy hands our hero is called upon once again...