Campus Grid Users
Here are a slection of case studies of Campus Grid users; a wider selection of e-Research projects are documented through the presentations given at the various e-Research workshops. The slides from these meetings can be found on the appropriate workshop webpages:
- e-Research Open Meeting 2009 (Campus Grid and Thames Blue users)
- Introduction to e-Research in Arts and Humanities Workshop (Arts and Humanities projects)
- e-Research Open Meeting - June 2008 (Services overview)
Brain Computer Interfaces, Ian Daly, SSE
Brain Computer Interfaces (BCIs) provide an alternative means for control of a computer by bypassing the peripheral nervous system. Unlike traditional methods of computer control, such as the mouse or keyboard, a BCI allows the user to control a computer via thought alone. Thus BCIs allow people with motor disabilities, such as tetraplegia, control of their environment and offer them additional channels of communication.
BCIs have also being used in the control of other systems such as spelling and typing programs, prosthetics limbs and even entertainment systems such as new types of computer games.
BCIs use signal processing methods to translate signals recorded from the human brain into control signals for some aspect of a computer. Our research concentrates on finding good signal types and the signal processing algorithms to best extract, process and classify these signals.
The complexity of the human brain means that many different types of signals can potentially be recorded from the brain. Finding the features that best describe these signals often involves a very large number of calculations. To run these calculation within an reasonable amount of time Matlab programs are compiled and run on the Campus Grid system Without the Campus Grid many of the processing methods we employ would take many days to run. The Campus Grid allows us to calculate results several hundred times faster then would otherwise be possible. These results then allow us to find new ways to control a computer via neurological activity and in the process reveal new and exciting information about the way our brains works.
How good is your Weather Forecast? - Lizzie Froude, ESSC
Whether we want to have a picnic next week or want to know where to go on our summer holidays, we all know that weather forecasting is never as good as we would like it to be. But when it comes down to predicting where tropical storms will strike the accuracy becomes much more important.
As the atmosphere is a chaotic system, modern medium-range weather forecasting uses "Ensemble Prediction". The computer model used to forecast the weather is run many times with tiny variations in the input data, which is based on the current observations. If all the runs produce nearly the same results then the forecasters can be much more confident about the forecast.
Lizzie Froude has been using the TRACK storm tracking software to compare the paths of storms in the real world with the predictions made by the weather forecasts. This work has shown where the various models have strengths and weaknesses and confirmed that the Ensemble Prediction method is better than running the model once.
To run the ensemble prediction requires many hundreds of small jobs to be run, just for one day's worth of forecasting.Therefore Lizzie has turned to the University of Reading Campus Grid which allows any researcher to run their computational experiments on library and lab computers when they would otherwise be unused.
The analysis in this work could potentially allow weather forecasters to understand the limitations of the prediction models that they use and to develop better weather forecasting models.
For more information on this work see: The Prediction of Extratropical Storm Tracks by the ECMWF and NCEP Ensemble Prediction Systems, 2007, L. S. R. Froude, L. Bengtsson and K. I. Hodges, Monthly Weather Review, 135, pp 2545-2567.
Face Recognition - Mian Zhou, SSE
Face recognition system are a holy grail in policing and security.A reliable system for face recognition could allow criminals to automatically be picked out from a crowd, which could greatly increase security at airports, train stations and sports games.Mian is investigating into one particular method for developing face recognition software using "Gabor Features".
Gabor features have already proved very effective for fingerprint and iris recognition, and this is not surprising as they extract information in a similar manner to mammal eyes. The advantage of applying these techniques to face recognition is that face recognition does not require the co-operation of the person being recognised, unlike iris and fingerprint recognition. Each Gabor feature has many parameters to describe it: position, orientation, area covered, size of feature, etc and so in his experiments Mian is employing 30,240 different Gabor features.
Due to the range of Gabor features there is some overlap in the information that makes up the different Gabor features.Knowing about this overlap or "mutual information" is important because mutual information represents wasted effort.With data on all the mutual information between each pair of features the set of features with the least mutual information can be chosen and the processing time minimised. Mian' is calculating this mutual information for each of the 457 million pairs of Gabor features. This requires a large number of calculations and would take 105 days on a desktop machine, but takes 20 hours on the Campus Grid.
The second part of Mian's work is to build up a system that can recognise one of 200 faces. Each of the individual Gabor features can be used as a way to tell whether a given picture is of the face that we are interested in or another face. On their own one of the features are not very good for this, but if the decision was made based on the most effective subset of the features it becomes very good at making the distinction. Mian is testing out an algorithm to construct these subsets for each of sample faces; on one machine this would take 2.1 years, but using the Campus Grid it only takes about a week.
For more information on this work see: Face Verification Using GaborWavelets and AdaBoost, 2006,Mian Zhou, Hong Wei, 18th International Conference on Pattern Recognition, (ICPR 2006).
vol. 1, pp 404 - 407
Climate Change -Kevin Hodges, ESSC
Kevin Hodges and his colleagues in the Environmental Systems Science Centre in Reading, and in Germany and Japan, have been trying to predict how the characteristics of storms will change with the changing climate. To do this they employed a climate model similar to those used to predict the weather for weather forecasts.
Initially the team validated that the climate model they were using realistically predicted the number of storms and their intensity in the current climate. To do this they ran the climate model and used the TRACK software to track the storms that were seen as the model was run over a 40 year period. The statistical distribution of these storms was found to agree well with what really happened over the same 40 year period.
The researchers than ran the model and TRACK software again, using the predictions for the future climate changes, to predict the same statistics about storms for the end of the 21st century. The two sets of statistics were then analysed, using Monte Carlo analysis, to see where there was a statically significant change in the storm statistics
The team used the University of Reading Campus Grid to both run the storm tracking software and perform the Monte Carlo analysis. The Campus Grid allows any researcher to run their computational experiments on library and lab computers when they would otherwise be unused.
While the team found that there was no change to the statistical distribution of storm intensities, nor was there a forecast for a greater number of storms in future, they did find that there was a reduction in weaker storms and an increase in the intensity of the stronger storms. The researchers also predict some localised changes for example: reduced precipitation in places like Australia, more storms north of the British Iles and changes in the characteristics of tropical storms.
For more information on this work see: Storm Tracks and Climate Change, 2006, L. Bengtsson, K. I. Hodges and E. Roeckner, Journal of Climate, V19, 3518-3543 and How may tropical cyclones change in a warmer climate?, 2007, L. Bengtsson, K. I. Hodges, M. Esch, N. Keenlyside, L. Kornblueh, J.-J. Luo and T. Yamagata, Tellus, V59A, pp 539-561.
The Certification of Endangered Mexican Cacti Project - Chris Yesson, Biological Sciences
The DEFRA-funded Certification of Endangered Mexican Cacti Project aims to support the conservation and sustainable harvest of Mexican desert cacti by providing molecular tools which can be used to identify plants to species, to determine their parentage and to locate the populations that they were collected from originally.
We have collected DNA barcodes for more than 600 Mexican Cacti and have used computational methods (phylogenetic analysis using MrBayes) to understand the relationships between species. Using the National Grid Service (NGS), of which Reading is an Affiliate, dramatically reduced the time of analysis and permitted a more thorough exploration of our data.
For more details see the project website http://www.uaq.mx/ccma.