School of Computing Science

Events

Students sitting in a lecture theatre

Explore upcoming seminars, guest lectures, workshops, and other events hosted by the School of Computing Science.

Our events bring together students, researchers, industry partners, and the wider community to share ideas, showcase research, and foster collaboration.

Quick links

This Week’s EventsAll Upcoming EventsPast EventsWebapp

This Week’s Events

Section Strategy Meeting

Group: Formal Analysis, Theory and Algorithms (FATA)
Speaker: Various
Date: 05 October, 2026
Time: 15:00 - 16:00
Location: SAWB 423

Dense Retrieval within the Deadline: from Static to Load-sensitive Selective Dimension Pruning

Group: Information Retrieval (IR)
Speaker: Francesca Pezzuti, University of Pisa
Date: 05 October, 2026
Time: 15:00 - 16:00
Location: Sir Alwyn Williams Building, 422 Seminar Room

Bio:
Francesca Pezzuti is a PhD student in Information Engineering at the University of Pisa. Her research asks how search engines should evolve as large language models act both as core search engine components and as autonomous users. Spanning neural web crawling, adaptive dense retrieval, and agentic search behaviour, her work seeks to balance search effectiveness with computational efficiency and scalability.

Abstract:
Dense retrievers encode documents into high-dimensional embeddings, making similarity scoring costly under strict latency constraints. During traffic spikes, full-dimensional similarity scoring can saturate processing queues and cause queries to miss their deadlines. Static pruning reduces this cost by lowering embedding dimensionality offline but commits the system to a fixed efficiency-effectiveness trade-off. Load-sensitive selective pruning instead adapts embedding dimensionality at query time according to the current system load. Based on per-query deadlines, queue occupancy, and an offline-derived estimator of processing cost, it selects the largest dimensionality that fits a safe time budget. Results show that this strategy achieves a better effectiveness-throughput trade-off than both static pruning and full-dimensional scoring across varying query loads.

TBC

Group: Networked Systems Research Laboratory (NETLAB)
Speaker: TBC
Date: 08 October, 2026
Time: 10:00 - 11:00
Location: SAWB 423, Sir Alwyn Williams Building

Upcoming events

Section Strategy Meeting

Group: Formal Analysis, Theory and Algorithms (FATA)
Speaker: Various
Date: 05 October, 2026
Time: 15:00 - 16:00
Location: SAWB 423

Dense Retrieval within the Deadline: from Static to Load-sensitive Selective Dimension Pruning

Group: Information Retrieval (IR)
Speaker: Francesca Pezzuti, University of Pisa
Date: 05 October, 2026
Time: 15:00 - 16:00
Location: Sir Alwyn Williams Building, 422 Seminar Room

Bio:
Francesca Pezzuti is a PhD student in Information Engineering at the University of Pisa. Her research asks how search engines should evolve as large language models act both as core search engine components and as autonomous users. Spanning neural web crawling, adaptive dense retrieval, and agentic search behaviour, her work seeks to balance search effectiveness with computational efficiency and scalability.

Abstract:
Dense retrievers encode documents into high-dimensional embeddings, making similarity scoring costly under strict latency constraints. During traffic spikes, full-dimensional similarity scoring can saturate processing queues and cause queries to miss their deadlines. Static pruning reduces this cost by lowering embedding dimensionality offline but commits the system to a fixed efficiency-effectiveness trade-off. Load-sensitive selective pruning instead adapts embedding dimensionality at query time according to the current system load. Based on per-query deadlines, queue occupancy, and an offline-derived estimator of processing cost, it selects the largest dimensionality that fits a safe time budget. Results show that this strategy achieves a better effectiveness-throughput trade-off than both static pruning and full-dimensional scoring across varying query loads.

TBC

Group: Networked Systems Research Laboratory (NETLAB)
Speaker: TBC
Date: 08 October, 2026
Time: 10:00 - 11:00
Location: SAWB 423, Sir Alwyn Williams Building

Single-Agent Stability in Hedonic Games with Constrained Coalition Sizes.

Group: Formal Analysis, Theory and Algorithms (FATA)
Speaker: Adam Dunajski, University of Edinburgh
Date: 13 October, 2026
Time: 15:00 - 16:00
Location: 422 Sir Alwyn Williams (SAWB), University of Glasgow

Hedonic games are coalition formation games where agents form disjoint coalitions (groups) and each agent's utility depends only on the other agents within their group.

We study stability in additively separable hedonic games where coalition sizes have to respect a fixed lower and upper bound. We consider four classic notions of stability based on single-agent deviations, namely, Nash stability, individual stability, contractual Nash stability, and contractual individual stability.

For each stability notion, we consider two variants: in one, the coalition left behind by a deviator must still be of size at least the lower bound, and in the other there is no such constraint.

This talk will introduce the above model, and
provide a full picture of the existence of stable outcomes with respect to given parameters for the lower and upper bounds. Additionally, when there are only upper bounds on coalition sizes, we fully characterize the computational complexity of the associated existence problem, and for particular bounds we obtain polynomial-time algorithms for constructing stable outcomes.

This is joint work with Martin Bullinger, Edith Elkind, and Matan Gilboa, and appeared in the proceedings of SAGT.

Full paper available here: https://arxiv.org/abs/2510.12641

Containers for Dummies

Group: Formal Analysis, Theory and Algorithms (FATA)
Speaker: Jake Trevor, University of Glasgow
Date: 20 October, 2026
Time: 15:00 - 16:00
Location: 422 Sir Alwyn Williams (SAWB), University of Glasgow

Recursively defined structures are ubiquitous in computer science. When formalising such structures in a system like Lean or Agda, they are required to satisfy a condition called strict positivity. Work in such a system long enough, and you will surely run into positivity problems when trying to formalise things in the natural way. Containers are a tool developed (among other reasons) to tackle positivity problems. They are flexible enough to model a variety of structures in a way which is strictly positive by construction - and therefore, amenable to use in a theorem prover like lean or agda.

Despite their usefulness, containers remain something of a mystery to many people. Part of this, I believe, is due to the presentation in the existing work, which tends to be about the category theory underpinning them. In my experience however, they are much like monads; it is not necessary to understand the theoretical underpinning to use containers or find them useful. A simpler presentation is possible.

In this talk, I will give this simpler presentation. I will explain what containers are, why they are useful, and how to use them, without reference to the categorical
underpinnings. This will be heavily motivated by examples, which have been adapted from my own research work.

TBD

Group: Networked Systems Research Laboratory (NETLAB)
Speaker: TBD
Date: 22 October, 2026
Time: 10:00 - 11:00
Location: SAWB 423, Sir Alwyn Williams Building

TBD

How User-AI Mistreatment Occurs and Matters in Conversational Systems?

Group: Systems Seminars
Speaker: Fanqi Zeng, University of Oxford
Date: 03 November, 2026
Time: 14:00 - 15:00
Location: Room 422, Sir Alwyn Williams Building and Teams

Abstract:
Safety research often focuses on model- generated harms, but users may also direct hostility, coercion, and adversarial pressure at models. Understanding how and when that occurs is essential for accurately interpreting model behaviour, alignment drift, and real-world deployment risks. In this paper, we audit 777K English LMSYS-Chat-1M conversations with two independent detectors: an eight-category lexicon for hostility directed at the model, and the dataset's moderation signal; and show that they capture different, weakly overlapping phenomena. The lexicon identifies insults, threats, and jailbreak coercion aimed at the assistant, while moderation flags are dominated by toxic-content solicitation rather than hostility at the model. Together, they mark about 5% of user turns; adjusting the narrower lexicon-harassment union for measured precision puts mistreatment aimed at the assistant at 0.90%. These absolute rates describe arena-style evaluation traffic and should not be read as deployment-wide base rates. We find that user hostility varies 13-fold across models, driven largely by who each model attracts rather than by model behaviour: first-turn hostility spreads far wider than post-response hostility, and more than fifteenfold separates the extremes even after deduplicating opening prompts. Within conversations, assistant apologies are consistently associated with higher odds of next-turn hostility under both detectors; the effect survives restricting to non-refused prior turns and to jailbreak-free conversations, and is positive in 20 of 23 models. Yet across models, more apologetic models receive less hostility overall. Finally, hostility also shows temporal structure, with coercive openings front-loading the first turn while affective hostility accumulates over a session. We release the lexicon, the detector cross-validation pipeline, and all derived tables.
 
Speaker's bio:
Dr Fanqi Zeng is a UKRI Metascience AI Early Career Fellow (Principal Investigator) in the Department of Sociology and a Junior Research Fellow at Wolfson College, University of Oxford. He is also a Research Associate at the Oxford Internet Institute and a Research Fellow at the Institute for New Economic Thinking at Oxford. He holds a PhD in Engineering Mathematics from the University of Bristol. His research bridges AI and the social sciences, harnessing computational methods and large-scale data to illuminate patterns within complex societal and natural systems. His current work explores the transformative impact of AI on society and human behaviour. His work has been funded by UKRI and the John Fell Fund. His research has appeared in leading interdisciplinary journals, including Science, Nature Cities, Nature Communications, PNAS, and PNAS Nexus.

"Sticking their heads out above the parapets": Lived experiences of legal risks in research

Group: Systems Seminars
Speaker: Daniel Thomas, University of Strathclyde
Date: 24 November, 2026
Time: 14:00 - 15:00
Location: Room 422, Sir Alwyn Williams Building and Teams

Overbroad computer crime, intellectual property, and other laws are well known to create legal risks that can discourage essential research. Notable examples include the US Computer Fraud and Abuse Act and the UK Computer Misuse Act. Because such laws fail to distinguish malicious hacking from good-faith testing and research, researchers face serious legal risks for public-interest research activity like identifying software or hardware vulnerabilities or scraping data. Despite the research community’s broad awareness of these risks, our understanding of their practical impacts is limited, as most of the community’s knowledge comes from anecdotal evidence rather than systematic study. We conduct the first qualitative study focused on researchers’ lived experiences, to empirically document the impacts of legal risks and threats on research and researchers, and how researchers navigate legal risk situations. Our study engages two participant groups: researchers with legal-risk experiences in the UK or the US (NR = 36), who discuss 130 projects and incidents spanning over three decades, and professionals that offer support to researchers navigating legal risks (NS = 8), who have collectively supported thousands of researchers. We thus provide an unprecedented big-picture view of researchers’ experiences with legal risks. We synthesise actionable strategies for researchers, and our findings provide evidence to support policy reform.

Past events

To view past events, please click here

Events Webapp