School Seminar Series

School Seminar – Peter Macgregor “Fast Dynamic Algorithms for Modern Clustering”

You are warmly invited to the second School Seminar:

Speaker: Peter Macgregor

Title: Fast Dynamic Algorithms for Modern Clustering

Abstract: Spectral clustering and DBSCAN both have long histories as theoretically grounded, general-purpose clustering algorithms. However, they face practical challenges when scaling to large datasets which have limited their adoption in practice.

In recent work, we have developed several improvements to these algorithms which improve their running time and space complexity while preserving their performance guarantees and generalising them to dynamically changing datasets. We make use of several algorithmic techniques including sparsification, dimensionality reduction, and random sampling. In this talk, I will present the recent progress and make the case that it’s time to challenge k-means’ dominance as the ‘default’ clustering algorithm.

Date & Time: Thursday 16/10/2025 11am-12pm

Location: JC 1.33A

Please do come along and join us! 🙂

School Seminar – Ian Gent, “How Not To Do It”

You are warmly invited to the next School Seminar:

Speaker: Ian Gent

Title: How Not To Do It

Abstract: Empirical methods are a vital part of a researcher’s toolbox. Which means that the more senior a researcher is, the more tools they have dropped on their feet!  I will share real mistakes which I or my colleagues made in analysing SAT and CP algorithms, and which we are prepared to own up to! Hopefully, you can learn from our mistakes instead of being doomed to repeat them.  As an old academic I’ll also take the chance to offer some advice on being an academic. Like most advice given by old people this advice is valued principally by the person giving it and may be worthless to anyone else.

Date & Time: Tuesday 30/09/2025 11am-12pm

Location: JC 1.33A

Please do come along and join us 🙂

 

PGR Seminar with Gen Li + Jess McGowan

The next PGR seminar is taking place this Friday 2nd May at 2PM in JC 1.33a

Below are the Titles and Abstracts for Gen and Jess’ talks – Please do come along if you are able.

Gen Li

Title: Visualization of clinical pathways based on sepsis comorbidities

Abstract: Sepsis is a severe infectious syndrome that can lead to critical illness and death. At present, most retrospective studies on sepsis focus on diagnosis and mortality risk prediction, with relatively limited attention to patients’ medical backgrounds. Comorbidities, as an important factor affecting the severity of the disease and treatment outcomes, present complex and variable characteristics in the treatment process. However, current research in this field generally lacks in-depth analysis of clinical pathways such as patient transfers and treatment interventions during hospitalization, which limits the development of personalized treatment strategies. Based on this, our research plans to use machine learning methods to extract similar comorbidity sub-groups of sepsis patients from electronic health records (EHRs), and further combine them with advanced visualization technology to explore the clinical pathways of these sub-groups. The research aims to help clinicians gain insights into the potential relationship between sepsis and related comorbidities, improve the interpretability of patients’ clinical records, and thus develop more effective treatment and management strategies for patients.

Jess McGowan

Title: Roll For Initiative: From Play to Personas

Abstract: In user centred design, designing for a wide target audience can lead to systems attempting to please everyone and thus pleasing no-one. Using a persona, i.e. a single member of that target audience, and designing a system dedicated to their needs results in a more focused design, which leads to improved usability. However, the design of personas is largely unstructured, with no clearly agreed methodology behind their creation. The solution to this could be found in Tabletop Role Playing Games (TTRPGs), which tend to feature clearly structured character creation instructions. This project aims to investigate to what extent can TTRPG character creation instructions aid the design of personas.

SICSA DVF Seminar – Dr André G. Pereira

We had our first School seminar of the semester today. The speaker was André G. Pereira visiting Scotland on a SICSA DVF Fellowship. André is working on AI Planning problems, an area that is closely related to the work of our own Constraint Programming research group.

Title: Understanding Neuro-Symbolic Planning

Abstract: In this seminar, we present the area of neuro-symbolic planning, introducing fundamental concepts and applications. We focus on presenting recent research on the problem of learning heuristic functions with machine learning techniques. We discuss the distinctions and particularities between the “model-based” and “model-free” approaches, and the different methods to address the problem. Then, we focus on explaining the behavior of “model-free” approaches. We discuss the generation of the training set, and present sampling algorithms and techniques to improve the quality of the training set. We also discuss how the distribution of samples over the state space of a task, together with the quality of its estimators, are directly related to the quality of the learned heuristic function. Finally, we empirically detail which factors have the greatest impact on the quality of the learned heuristic function.

Biography: Dr. André G. Pereira is a professor at the Federal University of Rio Grande do Sul, Brazil. His research aims to develop and explain the behavior of intelligent systems for sequential decision-making problems. Dr. Pereira has authored several papers on top-tier venues such as IJCAI, AAAI, and ICAPS. These papers contribute towards explaining the behavior of heuristic search algorithms, how to use combinatorial optimization-based reasoning to solve planning tasks, and how to use machine learning techniques to produce heuristic functions. Dr. Pereira is a program committee member of IJCAI and AAAI. His doctoral dissertation was awarded second place in the national Doctoral Dissertation Contest on Computer Science (2017), and first place in the national Doctoral Dissertation Contest on Artificial Intelligence (2018). Dr. Pereira advised three awarded students on national events, including first place and finalist in the Scientific Initiation Work Contest (2018, 2022), and finalist in the Master Dissertation Contest on Artificial Intelligence (2020).

Seminar – Richard Connor – 5th November

The second school seminar on 5th November at 2pm, on Teams.  If you do not have the Teams link available please contact the organiser, Ian Gent.

Dimensionality Reduction in non-Euclidean Spaces
Richard Connor
Deep Learning (ie 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 these 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 existing dimensionality reduction techniques perform analysis over the coordinate space 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.
This is work in progress; anyone interested is more than welcome to collaborate!

Georgios Gerasimou (University of St Andrews): Frontiers in computational revealed preference analysis

RESCHEDULED: please note the changed date and a non-standard time!

Abstract: Prest is a recently published piece of open-source software for computational revealed preference analysis that provides novel ways to estimate decision makers’ preferences over choice alternatives by analysing their observable choice behaviour. This software is informed by classic as well as recent developments in economic revealed preference theory. Some of the recent developments take the form of models that are computationally complex. This complexity currently hinders the inclusion of these models in the Prest toolkit. The presentation will first aim to describe the primary ideas underpinning Prest and illustrate them with examples from its existing toolkit. It will then proceed with a discussion of some of the challenges pertaining to the expansion of that toolkit with more models and operations. The presentation will be self-contained and no prior background in economics will be necessary.

Speaker Bio: Georgios is a Reader in Economics at the University of St Andrews, working mainly on decision theory and revealed preference analysis. In the latter research programme, Georgios’ work aims to improve our understanding of people’s decision processes and preferences through theoretical, experimental/empirical as well as computational methods. Georgios co-developed the Prest software program for computational revealed preference analysis (https://prestsoftware.com/).

Event details

  • When: 17th February 2020 14:00 - 15:00
  • Where: Cole 1.33b
  • Series: School Seminar Series
  • Format: Seminar

Philippe Palanque (University of Toulouse): Harnessing Usability, UX and Dependability for Interactions in Safety Critical Contexts

Abstract: Innovation and creativity are the research drivers of the Human-Computer Interaction (HCI) community which is currently investing a vast amount of resources in the design and evaluation of “new” user interfaces and interaction techniques, leaving the correct functioning of these interfaces at the discretion of the helpless developers. In the area of formal methods and dependable systems the emphasis is usually put on the correct functioning of the system leaving its usability to secondary-level concerns (if at all addressed). However, designing interactive systems requires blending knowledge from these domains in order to provide operators with enjoyable, usable and dependable systems. The talk will present possible research directions and their benefits for combining several complementary approaches to engineer interactive critical systems. Due to their specificities, addressing this problem requires the definition of methods, notations, processes and tools to go from early informal requirements to deployed and maintained operational interactive systems. The presentation will highlight the benefits of (and the need for) an integrated framework for the iterative design of operators’ procedures and tasks, training material and the interactive system itself. The emphasis will be on interaction techniques specification and validation as their design is usually the main concern of HCI conferences. A specific focus will be on automation that is widely integrated in interactive systems both at interaction techniques level and at application level. Examples will be taken from interactive cockpits on large civil commercial aircrafts (such as the A380), satellite ground segment application and Air Traffic Control workstations.

Speaker Bio: Dr. Philippe Palanque is Professor in Computer Science at the University Toulouse 3 “Paul Sabatier” and is head of the Interactive Critical Systems group at the Institut de Recherche en Informatique de Toulouse (IRIT) in France. Since the late 80s he has been working on the development and application of formal description techniques for interactive system. He has worked for more than 10 years on research projects to improve interactive Ground Segment Systems at the Centre National d’Etudes Spatiales (CNES) and is also involved in the development of software architectures and user interface modeling for interactive cockpits in large civil aircraft (funded by Airbus). He was involved in the research network HALA! (Higher Automation Levels in Aviation) funded by SESAR programme which targets at building the future European air traffic management system. The main driver of Philippe’s research over the last 20 years has been to address in an even way Usability, Safety and Dependability in order to build trustable safety critical interactive systems. He is the secretary of the IFIP Working group 13.5 on Resilience, Reliability, Safety and Human Error in System Development, was steering committee chair of the CHI conference series at ACM SIGCHI and chair of the IFIP Technical Committee 13 on Human-Computer Interaction.

 

Event details

  • When: 3rd February 2020 11:00 - 12:00
  • Where: Cole 1.33a
  • Series: SACHI Seminar Series, School Seminar Series
  • Format: Seminar

Nguyen Dang (University of St Andrews): Hyper-Parameter Tuning for an evolutionary algorithm

Abstract: In this talk, I will present a case study to illustrate how automated algorithm configuration can be used to gain insights into theoretical results on an evolutionary algorithm, namely the (1+(λ,λ)) Genetic Algorithm. This work is a collaboration with Carola Doerr.

The (1+(λ,λ)) Genetic Algorithm is an evolutionary algorithm that has interesting theoretical properties. It is the first algorithm where the benefit of crossover operator is rigorously proved. It is also the first example where self-adjusting parameter choice is proved to outperform any static parameter choice. However, it is not very well understood how the hyper-parameter settings influences the overall performance of the algorithm. Analyzing such multi-dimensional dependencies precisely is at the edge of what running time analysis can offer. In this work, we make a step forward on this question by presenting an in-depth study of the algorithm’s hyper-parameters using techniques in automated algorithm configuration.

Speaker bio: Dr Nguyen Dang is a post-doc in the Constraint Programming group at the University of St Andrews. Her main research focus is on automated algorithm configuration, algorithm selection and their applications in various contexts. These techniques make use of statistical methods and machine learning for fine-tuning of algorithm parameters, assessing parameters’ importance and building algorithm portfolios. Another line of her research is about solving combinatorial optimisation problems using metaheuristic algorithms.

Event details

  • When: 11th February 2020 14:00 - 15:00
  • Where: Cole 1.33b
  • Series: School Seminar Series
  • Format: Seminar

Matt Blackledge (Institute of Cancer Research): Clinical Computational Imaging: Perspectives in Oncology

Abstract: There is an ever-increasing burden on imaging departments to deliver high-throughput assessment of medical images.  MRI in particular provides the advantage of full-body coverage and and a variety of quantitative imaging techniques, such as diffusion-weighted MRI, that can offer potent biomarkers for disease response and prognosis; with the advent of accelerated imaging techniques, many quantitative images can now be acquired in a single patient scan.  Increases in computational power and the advent of methodologies such as deep-learning may help to deliver on the promise of truly personalised, image-guided therapies; by helping clinicians to better understand the complexities within multi-parametric MRI it may be possible to derive a non-invasive “digital biopsy” that can be monitored during treatment.  In this presentation, we will review recent developments within the Computational Imaging group at the Institute of Cancer Research and Royal Marsden Hospital, demonstrating how novel algorithms and deep learning can be used to assist in the response assessment of advanced prostate cancer and soft-tissue sarcoma.

Speaker Bio: Dr Matt Blackledge is the team-leader of computational imaging at the ICR, where he has been developing computational techniques in MRI for over a decade.  He is funded by both CRUK and Sarcoma UK to innovate novel approaches to image analysis from MRI and X-ray CT to improve cancer patient outcomes in a variety of disease types.  He is particularly interested in how AI can be used to (i) further accelerate MR-image acquisition, (ii) understand cancer heterogeneity in images, and (iii) probe the link between quantitative imaging biomarkers and their underlying biology.

 

Event details

  • When: 4th February 2020 14:00 - 15:00
  • Where: Cole 1.33b
  • Series: School Seminar Series
  • Format: Seminar