School of Computing Science

Learned Sparse Retrieval for Long Documents

Group: Information Retrieval (IR)
Speaker: Emmanouil Georgios Lionis, University of Glasgow
Date: 17 November, 2025
Time: 15:00 - 16:00
Location: Sir Alwyn Williams Building, 422 Seminar Room

Title:
Learned Sparse Retrieval for Long Documents

Abstract:
Retrieving relevant information from large text collections remains a core challenge in Information Retrieval (IR). Recent advances in encoder-only models have driven progress in Learned Sparse Retrieval (LSR), which represents queries and documents as sparse vectors to enable efficient matching. However, these models often struggle with long documents due to their limited context windows, leading to incomplete representations and potential retrieval gaps. In this talk, we will explore recent techniques for extending Learned Sparse Retrieval to handle long documents effectively. We will discuss current state-of-the-art approaches, analyse their strengths and limitations, and highlight open challenges and opportunities for future research in this space.

Bio:
Emmanouil Georgios (Akis) Lionis is a first-year PhD student at the University of Glasgow, supervised by Dr. Debasis Ganguly and Dr. Sean MacAvaney. His research focuses on advancing neural retrieval systems that scale efficiently without compromising retrieval quality. He graduated with honours from the University of Amsterdam, where he was part of the ELLIS MSc Honours program, connecting outstanding students with leading researchers across Europe. During his master’s thesis, he conducted an internship at TKH AI, allowing him to combine academic research with industrial applications.

xMem: A CPU-Based Approach for Accurate GPU Memory Prediction in Deep Learning Training

Group: Systems Seminars
Speaker: Jiabo Shi, University of Glasgow
Date: 18 November, 2025
Time: 14:00 - 15:00
Location: Room 422, Sir Alwyn Williams Building and Zoom

The widespread adoption of Deep Learning (DL) in diverse application areas has significantly increased the demand for GPUs. Consequently, GPU resources are scarce and are managed in clusters to maximize resource utilization. However, this shift introduces new debugging challenges when training DL models on shared clusters particularly Out-Of-Memory (OOM) errors, an issue commonly reported in industry and academic literature. Existing solutions for avoiding OOM primarily rely on static analysis of the DL model’s computational graph, or leverage GPU resources directly or indirectly to estimate the peak memory required for training the given task on the target GPU. Unfortunately, relying on GPUs for these predictions exacerbates resource contention and increases scheduling challenges. Furthermore, the dynamic nature of model development limits the accuracy of static analysis to estimate peak memory usage. To address these limitations, we propose xMem, a novel tool that uses CPU-based analysis to accurately predict the memory required for model training on a GPU. By eliminating the reliance on GPUs for memory estimation, xMem promotes efficient GPU utilization while mitigating OOM errors. Our empirical evaluation of 16 DL models (a total of 5,040 runs) demonstrates that, compared to state-of-the-art GPU memory estimators, xMem decreases the median relative error by 84.32%, reduces the average probability of estimation failure by 73.44%, accelerates the runtime by 50.16%, and improves memory conservation by 125.36%.

About the Speaker:

Jiabo Shi is a PhD student at the University of Glasgow, supervised by Dr Yehia Elkhatib and Professor Dimitrios Pezaros. His interests focus on machine learning performance measurement and prediction, as well as developing efficient and reliable resource scheduling systems for deep learning clusters. His current PhD research investigates the widespread resource waste in large-scale deep learning clusters, with a focus on job failures due to Out-of-Memory (OOM) errors amid the global GPU shortage. 

[FATA Seminar] TBA

Group: Formal Analysis, Theory and Algorithms (FATA)
Speaker: Alceste Scalas, DTU Compute - TU Denmark
Date: 18 November, 2025
Time: 15:00 - 16:00
Location: Room 422, SAWB

TBA

TBA

Group: Programming Languages at University of Glasgow (PLUG)
Speaker: Anton Lorenzen, University of Edinburgh
Date: 19 November, 2025
Time: 15:00 - 16:00
Location: F121 Lilybank Gardens and Online

Anton Lorenzen from Edinburgh will give us a talk -- most likely on his work on Functional-but-in-place programming. Details TBA.

What makes a "serious" computer science department?

Group: School of Computing Science
Speaker: Stephen Kell, King's College London
Date: 21 November, 2025
Time: 16:00 - 17:00
Location: 422 Sir Alwyn Williams (SAWB), University of Glasgow

What are the hallmarks of "things done right" in a computer science 
department that aspires to world-class research and teaching? In this 
polemic I'll share my own thoughts on this question, based on 
experiences working at a few different departments (while enjoying a 
blissful absence of management responsibility), and invite thoughts from 
the audience. The struggle is real, in that no department does uniformly 
well by these criteria. I'll share some anecdotes about the obstacles 
that can arise and some possible tactics (or "armchair wisdom") that may 
sometimes help overcome them.

Learned Sparse Retrieval for Long Documents

Group: Information Retrieval (IR)
Speaker: Emmanouil Georgios Lionis, University of Glasgow
Date: 17 November, 2025
Time: 15:00 - 16:00
Location: Sir Alwyn Williams Building, 422 Seminar Room

Title:
Learned Sparse Retrieval for Long Documents

Abstract:
Retrieving relevant information from large text collections remains a core challenge in Information Retrieval (IR). Recent advances in encoder-only models have driven progress in Learned Sparse Retrieval (LSR), which represents queries and documents as sparse vectors to enable efficient matching. However, these models often struggle with long documents due to their limited context windows, leading to incomplete representations and potential retrieval gaps. In this talk, we will explore recent techniques for extending Learned Sparse Retrieval to handle long documents effectively. We will discuss current state-of-the-art approaches, analyse their strengths and limitations, and highlight open challenges and opportunities for future research in this space.

Bio:
Emmanouil Georgios (Akis) Lionis is a first-year PhD student at the University of Glasgow, supervised by Dr. Debasis Ganguly and Dr. Sean MacAvaney. His research focuses on advancing neural retrieval systems that scale efficiently without compromising retrieval quality. He graduated with honours from the University of Amsterdam, where he was part of the ELLIS MSc Honours program, connecting outstanding students with leading researchers across Europe. During his master’s thesis, he conducted an internship at TKH AI, allowing him to combine academic research with industrial applications.

xMem: A CPU-Based Approach for Accurate GPU Memory Prediction in Deep Learning Training

Group: Systems Seminars
Speaker: Jiabo Shi, University of Glasgow
Date: 18 November, 2025
Time: 14:00 - 15:00
Location: Room 422, Sir Alwyn Williams Building and Zoom

The widespread adoption of Deep Learning (DL) in diverse application areas has significantly increased the demand for GPUs. Consequently, GPU resources are scarce and are managed in clusters to maximize resource utilization. However, this shift introduces new debugging challenges when training DL models on shared clusters particularly Out-Of-Memory (OOM) errors, an issue commonly reported in industry and academic literature. Existing solutions for avoiding OOM primarily rely on static analysis of the DL model’s computational graph, or leverage GPU resources directly or indirectly to estimate the peak memory required for training the given task on the target GPU. Unfortunately, relying on GPUs for these predictions exacerbates resource contention and increases scheduling challenges. Furthermore, the dynamic nature of model development limits the accuracy of static analysis to estimate peak memory usage. To address these limitations, we propose xMem, a novel tool that uses CPU-based analysis to accurately predict the memory required for model training on a GPU. By eliminating the reliance on GPUs for memory estimation, xMem promotes efficient GPU utilization while mitigating OOM errors. Our empirical evaluation of 16 DL models (a total of 5,040 runs) demonstrates that, compared to state-of-the-art GPU memory estimators, xMem decreases the median relative error by 84.32%, reduces the average probability of estimation failure by 73.44%, accelerates the runtime by 50.16%, and improves memory conservation by 125.36%.

About the Speaker:

Jiabo Shi is a PhD student at the University of Glasgow, supervised by Dr Yehia Elkhatib and Professor Dimitrios Pezaros. His interests focus on machine learning performance measurement and prediction, as well as developing efficient and reliable resource scheduling systems for deep learning clusters. His current PhD research investigates the widespread resource waste in large-scale deep learning clusters, with a focus on job failures due to Out-of-Memory (OOM) errors amid the global GPU shortage. 

[FATA Seminar] TBA

Group: Formal Analysis, Theory and Algorithms (FATA)
Speaker: Alceste Scalas, DTU Compute - TU Denmark
Date: 18 November, 2025
Time: 15:00 - 16:00
Location: Room 422, SAWB

TBA

TBA

Group: Programming Languages at University of Glasgow (PLUG)
Speaker: Anton Lorenzen, University of Edinburgh
Date: 19 November, 2025
Time: 15:00 - 16:00
Location: F121 Lilybank Gardens and Online

Anton Lorenzen from Edinburgh will give us a talk -- most likely on his work on Functional-but-in-place programming. Details TBA.

What makes a "serious" computer science department?

Group: School of Computing Science
Speaker: Stephen Kell, King's College London
Date: 21 November, 2025
Time: 16:00 - 17:00
Location: 422 Sir Alwyn Williams (SAWB), University of Glasgow

What are the hallmarks of "things done right" in a computer science 
department that aspires to world-class research and teaching? In this 
polemic I'll share my own thoughts on this question, based on 
experiences working at a few different departments (while enjoying a 
blissful absence of management responsibility), and invite thoughts from 
the audience. The struggle is real, in that no department does uniformly 
well by these criteria. I'll share some anecdotes about the obstacles 
that can arise and some possible tactics (or "armchair wisdom") that may 
sometimes help overcome them.

Fuzzing Techniques for Automated Vulnerability Detection in IoT Firmware

Group: Systems Seminars
Speaker: Kai Feng, University of Glasgow
Date: 25 November, 2025
Time: 14:00 - 15:00
Location: Room 422, Sir Alwyn Williams Building and Zoom

Security flaws in microcontroller (MCU) firmware are devastating, as patching them in the field is often impractical or impossible. Unfortunately, standard security testing techniques like fuzzing are slow and inaccurate on embedded systems due to their tight coupling with specialized hardware. This talk presents some new testing strategies that moves from slow emulation to testing on real hardware. We'll show how using data-flow guidance, instead of simple code coverage, more effectively finds critical bugs. Finally, we'll assess how new hardware features can be leveraged to proactively prevent entire classes of vulnerabilities.

[FATA Seminar] TBA

Group: Formal Analysis, Theory and Algorithms (FATA)
Speaker: Paul Besci, University of Oxford
Date: 25 November, 2025
Time: 15:00 - 16:00
Location: Room 422, SAWB

TBA

TBC

Group: Networked Systems Research Laboratory (NETLAB)
Speaker: Martin Nahalka
Date: 26 November, 2025
Time: 10:00 - 11:00
Location: Room 422, SAWB

TBC

Group: Systems Seminars
Speaker: Marc Juarez, University of Edinburgh
Date: 02 December, 2025
Time: 14:00 - 15:00
Location: Room 422, Sir Alwyn Williams Building and Zoom

Abstract:

TBC

Bio:

Marc Juarez is a Lecturer in Cyber Security and Privacy at the University of Edinburgh’s School of Informatics. Prior to his current appointment, he was a Postdoctoral Scholar in the Computer Science Department of the University of Southern California. His research focuses on investigating the privacy risks that arise from the application of machine learning techniques. More specifically, Marc’s work involves designing and evaluating countermeasures against machine learning-based attacks for privacy-aware Internet protocols, studying the privacy of deployed machine learning models, and developing mechanisms to measure the fairness properties of such models.

[FATA Seminar] TBA

Group: Formal Analysis, Theory and Algorithms (FATA)
Speaker: Josh Millar and Ryan Gibb, Imperial / Cambridge
Date: 02 December, 2025
Time: 15:00 - 16:00
Location: Room 422, SAWB

TBA

Title TBA

Group: Programming Languages at University of Glasgow (PLUG)
Speaker: Jacob Trevor
Date: 03 December, 2025
Time: 15:00 - 16:00
Location: F121 Lilybank Gardens and Online

Jake will give us a talk/rant about package managers. Details TBA.

TBC

Group: Systems Seminars
Speaker: Stephen McQuistin, University of St. Andrews
Date: 09 December, 2025
Time: 14:00 - 15:00
Location: Room 422, Sir Alwyn Williams Building and Zoom

TBC

Group: Networked Systems Research Laboratory (NETLAB)
Speaker: Kelsey Collington
Date: 10 December, 2025
Time: 10:00 - 11:00
Location: Room 422, SAWB

[FATA] Festive event

Group: Formal Analysis, Theory and Algorithms (FATA)
Speaker: N/A
Date: 16 December, 2025
Time: 14:00 - 16:00
Location: Room 422, SAWB

Calendar blocker

Measuring and understanding Distributed Denial of Service attacks

Group: Systems Seminars
Speaker: Daniel R. Thomas, University of Strathclyde
Date: 20 January, 2026
Time: 14:00 - 15:00
Location: Room 422, Sir Alwyn Williams Building and Zoom

Bio:

Dr Daniel R. Thomas is a Senior Lecturer at the University of Strathclyde where he is Director of the NCSC certified Academic Centre of Excellence in Cyber Security Research (ACE-CSR). His research interests are in measuring security and cybercrime so that we can monitor improvement, evaluate interventions and inform regulators. This reveals which techniques work and provides the missing economic incentives to improve security and reduce cybercrime. He co-organises the Strathclyde International Perspectives on Cybercrime Summer School [link](https://www.strath.ac.uk/science/computerinformationsciences/strathcyber/cybercrimesummerschool) , which next runs 24th-28th August 2026.

TBC

Group: Systems Seminars
Speaker: Tom Spink, University of St. Andrews
Date: 27 January, 2026
Time: 14:00 - 15:00
Location: Room 422, Sir Alwyn Williams Building and Zoom

TBC

TBC

Group: Systems Seminars
Speaker: Yuvraj Patel, University of Edinburgh
Date: 03 February, 2026
Time: 14:00 - 15:00
Location: Room 422, Sir Alwyn Williams Building and Zoom

TBC

TBC

Group: Systems Seminars
Speaker: Tao Chen, University of Birmingham
Date: 19 February, 2026
Time: 14:00 - 15:00
Location: Room 422, Sir Alwyn Williams Building and Zoom

TBC

HRI 2026

Group: Scottish Informatics and Computer Science Alliance (SICSA)
Speaker: SICSA Event, SICSA
Date: 16 March, 2026
Time: 00:00 - 00:00
Location: TBA

The ACM/IEEE International Conference on Human-Robot Interaction (HRI) is the premier venue for innovations on human-robot interaction. Sponsored by the ACM special interest groups on computer-human interaction (SIGCHI) and artificial intelligence (SIGAI) as well as the IEEE robotics and automation society (RAS), HRI brings together researchers spanning robotics, human-computer interaction, human factors, artificial intelligence, engineering, and social and behavioral sciences. The theme of the 21st edition of HRI is HRI Empowering Society. Our field has the potential to bring about positive change in many areas of our societies such as healthcare, transport, remote working, agriculture and industry. However, this change cannot happen if we do not engage properly with the end users who will potentially utilize robots in their jobs and daily lives. For this reason, HRI 2026 will focus on: 1) how we can ethically integrate robots in everyday processes without creating disruptions or inequalities, carefully thinking at the future of work and services; 2) how we can make them accessible to the general public (in terms of design, technical literacy and cost) with the final aim to make robots more willingly adopted as technological helpers. More information is available on the HRI 2026 website