Computing Science and Maths Seminars, 2019/2020
Seminars will take place in Room 4B96, Cottrell Building, University of Stirling. Normally, from 15.00 to 16.00 on Friday afternoons during semester time, unless otherwise stated. For instructions on how to get to the University, please look here.
Autumn 2019
| Date | Speaker | Title/Abstract |
|---|---|---|
| Friday 13th September |
Prof Richard Connor | Dimensionality Reduction in non-Euclidean Spaces Deep Learning (Convolutional Neural Networks) gives astoundingly good classification over many domains, notably images. Less well known, but perhaps more exciting, are similarity models that can be applied to their inner layers, where there lurk data representations that can give a much more generic notion of similarity. The problem is that the data representations are huge, and so searching a very large space for similar objects is inherently intractable. If we treat the data as high-dimensional vectors in Euclidean space, then a wealth of approximation techniques is available, most notably dimensionality reduction which can give much smaller forms of the data within acceptable error bounds. However, this data is not inherently a Euclidean space, and there are better ways of measuring similarity using more sophisticated metrics. The problem now is that all dimensionality reduction techniques perform analysis over the internal coordinates to achieve the size reduction. The more sophisticated metrics give only relative distances and are not amenable to analysis of the coordinates. In this talk, we show a novel technique which uses only the distances among whole objects to achieve a mapping into a low dimensional Euclidean space. As well as being applicable to non-Euclidean metrics, its performance over Euclidean spaces themselves is also interesting. |
| Thursday 19th September Note different day! |
Dr Ross Kelly, Department of Applied Mathematics, Liverpool John Moores University | Modelling changes in glutathione and mitochondrial superoxide homeostasis as a function of quinone redox metabolism Redox cycling is an understated mechanism of toxicity associated with a plethora of xenobiotics, responsible for preventing the effective treatment of serious conditions such as malaria and cardiomyopathy. Quinone compounds are notorious redox cyclers, present in drugs such as doxorubicin, which is used to treat a host of human cancers. However, the therapeutic index of doxorubicin is undermined by dose-dependent cardiotoxicity, which may be a function of futile redox cycling. In this study, a doxorubicin-specific in silico quinone redox metabolism model is described. Doxorubicin-GSH adduct formation kinetics are thermodynamically estimated from its reduction potential, while the remainder of the model is parameterised using oxygen consumption rate data, indicative of hydroquinone auto-oxidation. The model is then combined with a comprehensive glutathione metabolism model, facilitating the simulation of quinone redox cycling, and adduct-induced GSH depletion. Simulations suggest that glutathione pools are most sensitive to exposure duration at pharmacologically and supra-pharmacologically relevant doxorubicin concentrations. The model provides an alternative method of investigating and quantifying redox cycling induced oxidative stress, circumventing the experimental difficulties of measuring and tracking radical species. This in silico framework provides a platform from which GSH depletion can be explored as a function of a compound’s physicochemical properties |
| Friday 27th September |
Sarah Thomson | Anatomy of the Local Optima Level in Combinatorial Fitness Landscapes Interaction between optimisation problems and search algorithms affects the proficiency of search. Such interactions give rise to an object called a 'fitness landscape', which can be studied for meaningful structure. Topological features can be extracted. A local optima network (LON) is a compressed fitness landscape and is composed of local optima and their connectivity in terms of search transformations. Analysis of patterns and attributes of a LON can help with explanation, guidance or prediction for search algorithms. This talk describes how the local optima level in combinatorial optimisation problems relates to difficulty encountered of search algorithms. Features of local optima networks are proposed, extracted, and analysed. These are then used in algorithm performance prediction models. Methods for constructing LONs are also explored and compared. |
| Friday 4th October |
Dr Scott Denholm, SRUC | Star Tech, the final frontMIR: Utilising milk mid-infrared spectral data to predict hard-to-record phenotypes in dairy cows Today’s milk production systems are comprised of increasingly large herds of high-yielding dairy cattle, however, the number of on-farm staff has not increased proportionally. This has led to an increase in cow to staff ratios resulting in less attention available at the individual level. Maintaining a consistently high level of milk production whilst ensuring good animal health and welfare, as well as keeping financial loss to a minimum, is a delicate balance and one which is often difficult to uphold. Thus, new tools are required in order to assist farmers with the management of their herds. Mid infrared (MIR) spectroscopy of milk samples is an internationally recognised method for the routine collection of population-level data in large-scale milk recording. It is used to determine the fat and protein content of milk for the purpose of milk payments to dairy producers. This method of data collection is completely non-invasive and provides an exceptional quantitative analysis tool. Advances in knowledge, technology and methods have enabled researchers, milk recording companies and producers to utilise the by-product generated by MIR spectroscopy of milk samples (i.e., the spectral data) in order to routinely, and quickly, predict an increasing number of expensive and hard-to-record phenotypes. In this talk I will highlight the work of SRUC researchers working in this area, particularly some of our recent projects aimed at routinely predicting body energy, pregnancy and bovine tuberculosis in dairy cows. Speaker Bio: Scott Denholm is research scientist at Scotland’s Rural College working within the Animal & Veterinary Sciences research group. His research focuses on the genetics and genomics of immune-associated traits in dairy cows. Between 2009 and 2013 he was a Cefas funded maths PhD student at the University of Stirling investigating the long-term impact of Gyrodactylus salaris infections on UK Atlantic salmon populations (grant number FC1197). Until December 2016 he was a researcher on a BBSRC grant (BB/K002260/1) investigating the use of longitudinal data and a systems biology approach to understand and routinely predict health and welfare traits in dairy cattle. He is currently the main researcher on a BBSRC grant (BB/S009396/1) investigating the utility of deep learning techniques for the routine prediction of bovine tuberculosis status of dairy cows from mid infra-red spectral data. |
| 4pm, Monday 7th October LTV1 Note day and room! |
Dr Hannah Dee, Aberystwyth University | Ada Lovelace Day celebration Why Ada was awesome Ada Lovelace is often described as the world's first computer programmer, and has become an icon for women in science thanks to Ada Lovelace Day, the BCSWomen Lovelace Colloquium, the BCS Lovelace Medal and a number of other high profile events. But ... Who was she? Where did she come from? What exactly did she do? Was she a programmer in any sense? This talk will look at Ada in context, bringing together the personal and scientific stories behind a remarkable woman. Speaker Bio: Hannah is a Senior Lecturer at the Department of Computer Science, Aberystwyth University and has been on the national committee for BCSWomen for over a decade. In 2008, she founded the BCSWomen Lovelace Colloquium, the UK's premiere event for women computing students, and is now deputy chair. She is in the Computer Weekly women in tech "hall of fame", and is a Suffrage Science award holder. |
| 4pm, Thursday 10th October LTA5 Note day and room! |
Professor Vilmundur Gudnason, Icelandic Heart Association / University of Iceland | Fifty years of phenotyping in population studies of the Icelandic Heart Association The Icelandic Heart Association (IHA) is a non-profit institute that was founded in 1964 to battle cardiovascular illness in Iceland and is a recognized leader in the world-wide effort to discover and integrate scientific knowledge in order to enhance the quality of life for both young and old. For over 50 years, IHA has conducted large-scale studies of over 30,000 men and women born in Iceland between 1907 and 1935. The research has focused on the multiple causes of disability in old age including heart disease, high blood pressure, and Alzheimer's. The Reykjavik study started in 1967 and has since collected enormous amount of data, produced a multitude of scientific discoveries and numerous spin-off and follow-up research projects. One of the most prominent follow-ups is the Age, Gene/Environment Susceptibility (AGES Reykjavik) Study, a collaborative study between IHA, National Institute of Aging, and NIH. It was initiated to examine genetic susceptibility and gene/environment interaction as these contribute to phenotypes common in old age by sequencing the human genome and discovering candidate genes. AGES has sparked substantial innovation in the epidemiologic study of aging as well as identified novel opportunities to prevent diseases and limit disability. AGES has produced a host of widely published and highly cited scientific discoveries in top-tier journals, such as Nature, the Lancet, and New England Journal of Medicine. However, IHA is still evolving and looking for novel ways to exploit advancing technologies on its abundance of data for the benefit of healthy living for everyone. Speaker Bio: Professor Vilmundur Gudnason PhD, MD: Prof Gudnason has been the director of the Icelandic Heart Association Research Institute since 1999 and a professor of Cardiovascular Genetics at the University of Iceland since 1997. He is and has been the Principal Investigator of numerous large-scale research projects and consortia. He received his MD degree from the University of Iceland in 1985 and his PhD in genetics from University College London in 1995. Prof Gudnason is among the world's most highly cited scientist, with an h-index of 129 and appearing in two categories (Molecular Biology & Genetics and Clinical Medicine) on the current Clarivate list (formerly Thomson-Reuters) of highly cited researchers (top 1%). Prof Gudnason has extensively contributed to research on genetics and the epidemiology of cardiovascular diseases in the form of several book chapters and multiple high impact peer reviewed journals, including in Nature, Nature Genetics, New England Journal of Medicine, and the Lancet. |
| Friday 18th October |
Internal events | No seminar this week |
| Friday 25th October |
Richard Bradley | The Magic of Blockchain - Using Playing Cards An interactive workshop using cards and simple arithmetic to explore the basics of Blockchain. This workshop and presentation will introduce some of the technologies that combine to create Blockchain. In this session we will explore hash functions, encoding, sequencing, restoration, distributed cooperation, replication, integrity, and chaining. Using some basic rules that will evolve during the workshop, we will build an example Blockchain using playing cards. No knowledge of computing is required (though it helps). Warning: simple arithmetic will be performed (calculators are permitted). |
| Friday 1st November |
Dr. Julian Hall, University of Edinburgh | High performance solution of large-scale linear programming problems The requirement to solve large-scale linear programming (LP) problems is ubiquitous: they are the fundamental model in optimal decision-making, but also solved in vast numbers when finding the optimal solution of large-scale discrete optimization problems. Most LP problems are solved with the simplex algorithm, and this talk will focus on the computational challenges of using it to solve large-scale sparse problems on both serial and multi-core architectures. Performance is primarily determined by techniques to exploit the particular nature of the underlying numerical linear algebra requirements. These have characteristics that should be of interest to a general audience. There will also be an introduction to novel techniques for fast approximate solution of LP problems. Speaker Bio: Born and educated in Macclesfield, Dr. Julian Hall studied Maths at New College Oxford and then did a PhD at the University of Dundee under the supervision of Roger Fletcher FRS. His main research interest is in developing algorithmic and computational techniques for solving large scale linear programming (LP) problems on both serial and parallel computers. For many years this was focused on the revised simplex method but, recently, he has switched his attention to work which will lead to the development of novel algorithms. A consequential research interest is the application of these techniques in other areas of computational optimization and linear algebra. In collaboration with current and former PhD students, Dr. Hall is managing the development of the high performance open-source software linear optimization software HiGHS. |
| Friday 8th November |
Prof Annie Cuyt, University of Antwerp, Belgium | Exponential analysis in Computational Science and Engineering Multi-exponential analysis might sound remote, but it touches our lives in many surprising ways, even if most people are unaware of just how important it is. For example, a substantial amount of effort in the field of signal processing is essentially dedicated to the analysis of multi-exponential functions whose exponents are complex. The analysis of exponential functions whose exponents are very near each other is directly linked to superresolution imaging. As for multi-exponential functions with real exponents, they are used to portray relaxation, chemical reactions, radioactivity, heat transfer, fluid dynamics, to name just a few. Despite their computational efficiency and wide applicability, Fourier-based methods suffer some well-known limitations, such as a limited resolution and leakage in the frequency domain. They are also not well suited for aperiodic signals. Multi-exponential analysis on the other hand, may suffer from ill-posedness or ill-conditioning. The underlying method to reconstruct the exponential sum is already much older and goes back to de Prony. Both Fourier- and Prony-based methods make use of the Shannon-Nyquist theorem which states that the minimum sampling rate required to recover a signal is at least twice the maximum frequency present in the band-limited signal. Recent achievements, allowing to circumvent the Shannon-Nyquist sampling rate, to regularize the multi-exponential analysis and to break the curse of dimensionality in case of higher-dimensional analyses, have unlocked new potentials. Speaker Bio: Annie Cuyt is full professor at the Department of Mathematics and Computer Science of the University of Antwerp and is a lifetime member of the Flemish Royal Academy of Belgium for Science and the Arts since 2014. She received her Doctor Scientiae degree in 1982 from the same university, summa cum laude and with the felicitations of the jury. Subsequently she was a Research fellow with the Alexander von Humboldt Foundation (Germany) and she was honoured with a Masuda Research Grant (Japan). She is the author of more than 200 peer-reviewed publications in international journals and conference proceedings, the author or editor of several books and the organizer of a number of international events. Her current interests are in numerical approximation theory and its numerous applications in scientific computing. The latter vary from bioinformatics over different engineering applications (telecommunication, radar imaging, EM modeling, exponential analysis,...) to computer graphics. In view of her expertise she served on many national and international science foundation boards. |
| Friday 15th November |
Dr Jean Petric, Lancaster University | Which bugs are tests not finding? Software testing plays an important role in assuring the reliability of systems. In this work, we investigate how effective tests are at detecting different types of bugs and whether some types of bug evade tests more than others. We investigate seven bug types and analyse how often each goes undetected. Our results suggest that the bug detection rates of tests are relatively low, typically finding only about a half of all bugs. Some bug types are less well detected by tests than other bugs. Overall, we conclude that developers should not rely only on code coverage and mutation scores to measure the effectiveness of their tests. We uncover which bugs developers should test better and recommend ways in which tests can be improved. |
| Friday 22nd November |
Winter graduations | No seminar this week |
| Friday 29th November |
Available slot |
|
| Friday 6th December |
Georgi Tinchev, University of Oxford | Real-time LIDAR localization in natural and urban environments Localization is a key challenge in many robotics applications. In this work we explore LIDAR-based global localization in both urban and natural environments and develop a method suitable for online application. We present a method capable of achieving state-of-the-art performance while being three times faster than previous approaches, as well as occupying 70\% less memory without a significant loss of performance. Our approach leverages efficient deep learning architectures capable of learning compact point cloud descriptors directly from 3D data. The method uses an efficient feature space representation of a set of segmented point clouds to match between the current scene and the prior map. We show that down-sampling in the inner layers of the network can significantly reduce computation time without sacrificing performance. Our experiments demonstrate a factor of three reduction of computation with marginal loss in localization frequency. We evaluate the proposed methods on nine scenarios from six datasets varying between urban, park, forest and industrial environments. The proposed learning method can allow the full pipeline to run on robots with limited computation payload such as drones, quadrupeds or UGVs as it does not require a GPU at run time. |