Computing Science and Mathematics Seminars,
Autumn 2024
Unless otherwise state, seminars will take place in Room 4B96, Cottrell Building, University of Stirling from 13.00 to 14.00 on Friday afternoons, followed by informal discussions.
Autumn 2024
| Date | Speaker | Title/Abstract |
|---|---|---|
| Friday 11 October |
Wenyang Chu, Virtonomy GmbH, Munich, Germany | Accelerating Medical Device Development Through In-Silico Solutions This talk will explore the transformative potential of in silico (computer-based) simulations in medical device development, focusing on how advanced computational models are revolutionizing the design, testing, and validation of medical devices. Key topics:
Speaker bio: Wen-Yang Chu is the Chief Technology Officer and Co-Founder of Virtonomy.io, a pioneering company that applies digital twin technology to healthcare, specializing in both medical implant research and development as well as clinical trials. His work focuses on utilizing AI and biomedical simulations to create virtual patients, enabling the digitalization of clinical trials and accelerating medical device innovation. This innovative approach significantly reduces the reliance on animal testing and shortens timelines for regulatory approval. Virtonomy.io’s cloud-based platform, v-patients.com, integrates medical data, advanced simulations, and AI-driven models to optimize the design, testing, and validation of medical implants and devices. Wen-Yang holds a European Master of Science in Information and Communication Technologies from Universitat Politècnica de Catalunya and Université catholique de Louvain. His academic background in signal processing, AI, computer vision, and simulation technologies has equipped him with the expertise to drive groundbreaking innovations in healthcare. With over a decade of industrial experience in high-performance computing, machine learning, and software development, Wen-Yang has a proven track record of translating cutting-edge technologies into practical applications. Before founding Virtonomy.io, he worked as a Data Scientist and Software Architect at Philips, where he developed AI models and scalable image analysis systems for computational pathology. His contributions to medical device development earned him several Philips prestigious awards. Originally from Taiwan, Wen-Yang now lives in Belgium with his wife and two daughters, aged 10 and 13. In his spare time, he enjoys playing the violin, traveling, hiking, and spending quality time with his family. Committed to bridging the gap between academia and industry, Wen-Yang continues to push the boundaries of digital twin technology, advancing the development of next-generation medical devices. His interdisciplinary approach, combining AI, simulation, and real-world data, empowers medical device companies to test and validate implants more efficiently and precisely. |
| Monday 14 October 3-4pm 3A94 (different room!) |
Dalila Hamami, Mostaganem University, Algeria | Advancing Machine Learning and Artificial Intelligence for Healthcare and Education: An International Collaboration opportunity I will explore my current research projects leveraging machine learning and artificial intelligence to address pressing challenges in both healthcare and education. This talk will cover three significant projects:
By establishing links between Mostaganem University and Stirling university, this seminar serves as an invitation to UK academics and researchers who are interested in co-developing cutting-edge AI technologies, sharing resources, and contributing to impactful research in healthcare and education.
Speaker bio: Dr Dalila Hamami is a Senior Lecturer in Computing Science at Mostaganem University, Algeria. |
| Friday 18 October |
William Langdon, Department of Computer Science, University College London | Evolutionary Robustness Even in stable conditions biology can retain its ability for continued evolutionary improvement even after 75,000 generations. Instead of 36 years, with performance effectively exceeding a trillion GP operations per second, Genetic programming (GP) experiments can be run to a million generations in weeks on a single computer. Information theory explains why in small populations, GP populations converge and the rate of fitness improvement falls as huge GP trees become more robust to crossover. Mutation testing on C and C++ programs show that real software can also be robust to many source code changes. As with lisp functional language in tree GP, there is a tendency for deeply nested imperative code to be more robust. There are already examples of human written software systems that exceed a billion lines of (imperative) source code. Information theory's failed disruption propagation (FDP) helps to explain why maintaining, testing and debugging such deeply nested code repositories is hard and why software companies prefer unit testing of modules (each of which is typically only shallowly nested) rather than system testing of complete functional hierarchies. There is already SBSE work on automatically optimizing test oracles. FDP suggests systems should be built with many densely packed test agents so that disruption caused by bugs has little distance to travel before being discovered by an oracle. For evolutionary computing and artificial life experiments aiming for sustained innovation, we propose the use of "mangrove" architectures composed of many small trees which are intimate with their environment. For continuous innovative evolution the fitness function needs to be able to measure on average if genetic changes are good or not, or at least have made a difference. This means we must overcome robustness, without introducing chaos. We suggest this might be met by systems where the bulk of the code remains close to the fitness environment and the disruption caused by most mutations and crossovers has only a short depth to propagate in order to have a measurable fitness impact. Genetic programming and other types of Evolutionary Algorithms have long been demonstrated to be creative. A recently raised question was how much are they used? Data from the genetic programming bibliography for last year suggests 38 ±5% of published papers are primarily on applications which just happen to use GP. Many applications relate to health, civil engineering or solid state materials, e.g. batteries.
Speaker bio: Prof William B. Langdon is a leading researcher in the field of genetic programming and evolutionary computing. He is an honorary professor at the Department of Computer Science and a member of CREST (Centre for Research on Evolution, Search, and Testing) in the University College London. |
| Friday 25 October |
Reading week |
|
| Friday 1 November |
Elizabeth Wanner, School of Computer Science and Digital Technologies, Aston University | Mathematical models for Dominance Move: Comparison and complexity analysis Dominance move (DoM), a binary quality indicator, can be used in multi-objective and many-objective optimisation to compare two solution sets. DoM is very intuitive but hard to calculate due to its combinatorial nature. Different mathematical models are presented and analysed. A computationally fast approximate approach is also discussed. Computational results are promising and an upper bound analysis for the approximation ratio would be useful.
Speaker bio: Elizabeth Wanner received her B.S. degree in Mathematics in 1994, and her M.Sc. in 2002, both from the Universidade Federal de Minas Gerais, where she also completed her Ph.D. in Electrical Engineering in 2006. Until December 2022, she served as an Associate Professor in the Department of Computer Engineering at the Centro Federal de Educação Tecnológica de Minas Gerais, in Belo Horizonte, Brazil. Since January 2023, she has been a Reader in Computer Science at Aston University, Birmingham, UK. Her research focuses on evolutionary computation, global optimization, constraint handling, and multiobjective optimization. |
| Friday 8 November |
Yuanlin Gu, Computing Science and Mathematics, University of Stirling | The NARMAX Method: A Framework for System Identification & Interpretable Machine Learning The nonlinear autoregressive–moving-average with exogenous inputs (NARMAX) method provides a powerful tool for black-box system identification problems where the true model structure is unknown or hard to obtain. It have been applied to the identification of a wide range of nonlinear systems in various research fields including ecological, environmental, geophysical, societal and neurophysiological sciences. This talk begins with an introduction to the definition of system identification, followed by the classic NARMAX method and its associated term (feature) selection techniques, and then moves on to recent developments in interpretable machine learning and uncertainty analysis, with applications of multiple disciplines.
Speaker bio: Dr Yuanlin Gu received the B.S. degree in automatic control from Nanjing University of Aeronautics and Astronautics, China, the M.S. Degree in control systems and the PhD degree from University of Sheffield. He is a Lecturer in Computer Science with University of Stirling. Previously he has worked at University of Roehampton as a lecturer and Loughborough University as a researcher. His research interests include systems identification for complex nonlinear systems, interpretable machine learning, and uncertainty analysis. |
| Friday 15 November |
No seminar | |
| Friday 22 November |
No seminar | |
| Wednesday 27 November 12noon-1pm 4B96 same room different time |
GianCarlo Catalano, Computing Science and Mathematics, University of Stirling | Can we make sense of optimisation algorithms? The methods used to solve optimisation tasks can be very diversified, and modelling tasks can be thought of as a subset of them that is particularly successful. In the quest to find the optimal candidate for an objective, many different approaches have been developed to essentially make educated guesses towards an optimal result, but what they have in common is that they are hard to interpret for a variety of reasons. While there is lots of work into making AI models more transparent, there is still relatively little work which applies to optimisation algorithms in general, and I would argue that this is because they are so diversified that developing generalised approaches is tricky. This presentation will look into what makes optimisation methods so intractable, some ways in which this is being resolved, and some interesting thoughts about what it means to be “transparent”.
Speaker bio: GianCarlo is a second year MPhil student here at the University of Stirling, working on the ENDS (Explainability for Non-Deterministic Solvers) project being supervised by Sandy Brownlee and David Cairns. He completed his undergraduate degree in CS & Maths here at the University of Stirling in 2023, and developed an interest in Genetic Algorithms during the pandemic, which led him to make a dissertation about them and eventually into this research project. His work is funded by BT, who uses optimisation algorithms to decide the working patterns of their engineers but finds it difficult to have the results be used by the human managers (more details in the presentation). In order to alleviate this issue, they funded some research projects on how to make optimisation methods more transparent, such as GianCarlo’s work, which focuses on explaining individual results from metaheuristics so that they can be used with confidence by humans. |
| Friday 29 November |
Andrew Abel, Computer and Information Sciences, University of Strathclyde | Multimodal Communication: Machine Learning for Speech and Lipreading Speech and communication are multimodal, with mechanical and cognitive processing required for both the comprehension and production of speech. Speech is also heavily affected by environmental factors, with studies showing that as background noise increases, reliance on other factors also increases. This makes non-acoustic information very important for speech processing, particularly for those with hearing loss. One obvious solution to this is to make use of lipreading, particularly in very noisy environments, and in recent years, improvements in image processing and particularly with machine learning, have made this more feasible and more widely implemented. This seminar will introduce a number of speech phenomena, including the McGurk Effect and the Lombard Effect, and then present a number of recent developments in lipreading, both feature-based and end-to-end with deep learning models, and discuss how these can be applied and extended, as well as how these approaches can be examined and explained. However, a lot of work in the literature does not consider real world issues, such as how humans change their speech patterns in different environments, the availability of reliable visual data, and that even many of the state-of-the-art audio-only speech recognition models do not always generate trustworthy data, and these limitations will be explored.
Speaker bio: Andrew Abel is a Chancellors Fellow in the Dept of Computer and Information Sciences at Strathclyde University and received his Ph.D. from the University of Stirling in 2013, after conducting research into signal-image processing algorithms for enabling multi-modal speech processing technologies. Before Strathclyde, he worked at Anhui University in Hefei focusing on image processing, and was a lecturer at Xi'an Jiaotong-Liverpool University in China. His research interests are on the use of multiple modalities and machine learning to estimate and process speech and communication, encompassing both verbal and nonverbal communication. |
| Friday 6 December |
Nguyen Dang, School of Computer Science, University of St Andrews | Reinforcement Learning for Dynamic Algorithm Configuration Most algorithms have their own parameters that need to be tuned to achieve the best performance. In some cases, instead of finding the best static parameter setting for an algorithm, it is highly beneficial to adapt the parameter values while the algorithm is running. Dynamic Algorithm Configuration (DAC) focuses on developing techniques to solve this task in an automated and data-driven fashion. The aim is to learn a policy that maps from the current state of the algorithm to the best parameter value for that state during the solving process. DAC is an emerging topic and has a lot of potential applications in various domains. Given the dynamic nature of the task, Reinforcement Learning (RL) seems like a suitable family of techniques for tackling DAC problems. However, research on DAC methods is still in its early stage. It is not clear whether RL methods, which were originally developed for other domain applications such as robotics and game playing, are effective in DAC contexts. In this talk, I will give a brief introduction to DAC and present our recent study on benchmarking a commonly-used RL algorithm on DAC.
Speaker bio: Nguyen Dang is a Lecturer in Computer Science at the University of St Andrews. She received a PhD degree from KU Leuven (Belgium) in 2018 and was awarded a Leverhulme Early Career fellowship in 2020. Her research interests include automated algorithm configuration, automated algorithm selection, automated instance generation for benchmarking, and constraint programming. |
Previous Seminar Series
Computing Science and Mathematics
Faculty of Natural Sciences
University of Stirling
Stirling
FK9 4LA
Scotland
UK