Talk

PGR Seminar with Duong Phuc Tai Nguyen + Thomas Metcalfe

The next PGR seminar is taking place this Friday 9th May at 2PM in JC 1.33a

Below are the Titles and Abstracts for Duong and Tom’s talks – Please do come along if you are able.

Duong Phuc Tai Nguyen

Title: Enhancing Dynamic Algorithm Configuration via Theory-guided Benchmarks.

Abstract: Algorithms play a vital role in numerous domains, ranging from machine learning to optimization and simulation. Developing an algorithm typically requires making multiple design choices and fine-tuning parameters, a process that can be both labor-intensive and complex. This project seeks to automate this process by employing machine learning techniques, particularly deep reinforcement learning (deep-RL). By leveraging theoretical insights from evolutionary computation, we establish new benchmarks to assess RL methods for dynamic algorithm configuration and propose enhancement techniques to increase their effectiveness

Thomas Metcalfe

Title: Listening to Rhythms: Exploring Human-Phenology Attunement through Research Products and Decentralised Computing

Abstract: Humanity and the planet are in an epoch of ecological breakdown. Modern technological cultures have severed human awareness from the living rhythms of more-than-human worlds. This research explores how decentralised, situated technologies might foster embodied attunement between humans and the phenological rhythms of place.

This seminar is a work-in-progress. 7 months into his PhD, Tom will present his journey and current thinking on the foundations and potential direction of his project. You can expect to hear how he’s trying to shift research paradigms; the profound change in the project’s onto-epistemological perspectives; and how he hopes to make a contribution to the design and HCI communities.

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.

PGR Seminar with Dhananjay Saikumar

The next PGR seminar is taking place this Friday 25th April at 2PM in JC 1.33a

Below are the Title and Abstract for Dhananjay’s talk – Please do come along if you are able.

Title: Signal Collapse in One-Shot Pruning: When Sparse Models Fail to Distinguish Neural Representations

Abstract: The deep learning breakthrough in 2012, marked by AlexNet’s success on the ImageNet challenge, ushered in an era of increasingly large neural networks. Modern models now hold tens of millions to billions of parameters, enabling remarkable capabilities but creating serious challenges for deployment in real-world, resource-constrained environments. This has led to growing interest in model compression, with network pruning emerging as a widely adopted method to reduce computational and memory demands. Iterative pruning—based on repeated prune-retrain cycles—can retain accuracy but becomes infeasible at scale due to high computational cost. One-shot pruning, which removes parameters in a single step without retraining, offers a more scalable alternative but often results in severe accuracy degradation. For instance, pruning 80% of the parameters from RegNetX-32GF (a 100M+ parameter model) drops ImageNet accuracy from 80% to 1%, rendering the model unusable. This talk uncovers a new and fundamental bottleneck behind such failures: signal collapse, a previously overlooked phenomenon that disrupts the network’s ability to distinguish between inputs. To address this, a simple and efficient method called REFLOW is introduced, enabling sparse networks to recover strong performance without retraining or gradient computation. On RegNetX-32GF, REFLOW lifts accuracy from 1% to 73% at 80% sparsity—in under 15 seconds. These findings reframe the challenges of one-shot pruning and open new opportunities for practical and efficient deployment of deep learning models.

PGR Seminar with Leonid Nosovitsky + Xinya Gong

The next PGR seminar is taking place this Friday 11th April at 2PM in JC 1.33a

Below are the Titles and Abstracts for Leonid and Xinya’s talks – Please do come along if you are able.

Leonid Nosovitsky

Title: Bridging Theory and Practice: Advancing Multiparty Session Types for Industry Use

Abstract: Multiparty Session Types (MPST) provide a typing discipline for communication protocols. They allow to statically check that a code implementation conforms to a specified protocol; they can also verify that a protocol satisfies many safety properties like liveness and deadlock freedom, which are crucial in concurrent communicating systems.  Despite huge improvements in MPST research, different extensions have limitations.  For example, one of extensions is crash-handling. This branch was motivated by lack of network reliability to make MPST framework more applicable and usable in industrial scenarios. The crash-semantics theory introduced by Barwell et al. does not involve any constraints relaxations, which makes it very intricate for adoption in practical scenarios.  Our project concentrates on addressing usability limitations to make MPST more integrable into industrial applications.

Xinya Gong

Title and Abstract TBC

Distinguished Lecture Series 2025

This years Distinguished Lecture series was delivered yesterday ( Tuesday 1st April) by Professor Arthur Zimek, University of Southern Denmark in Odense, Denmark.

In his talk on, ‘Data Mining and the “Curse of Dimensionality”’ he considered the challenges of the “curse” from the perspective of data mining. In Talk 1, he discussed the “curse” in more detail, identifying relevant aspects or problems. In Talk 2, he considered clustering facing these problems and discussed some strategies and example methods for subspace clustering. In Talk 3, he discussed outlier detection, considering strategies for improved efficiency, effectiveness, and subspace outlier detection.

FactSet Talk – Insights into Prompt Engineering 25th March

On Tuesday the 25th of March, FactSet will be visiting the School of Computer Science. They will be doing a talk on Insights into Prompt Engineering, before hosting a networking and recruitment session with pizza. This is taking place in JC1.33 A/B from 3:00pm to 5:00pm. This is a free event to attend for all.

They are recruiting for their paid software engineer externship, paid software engineer internship and graduate software engineer roles. The externship is from the 7th to 19th July in London. This is a two-week program with hands-on experience working with Software Engineering teams. The software engineering summer internship is a 12-week program in summer, with interns joining an existing team at FactSet in London. The graduate program begins in September in London.

Hope to see you all there!

Fabrizio Capobianco (The Liquid Factory) Speaker on Friday 28th March

Speaker: Fabrizio Capobianco (Partner, The Liquid Factory)

Date: Friday 28th March

Time: 3:00

Venue: Jack Cole 1.33A/B

The Liquid Factory (www.theliquidfactory.com).

At The Liquid Factory, they support the next generation of European entrepreneurs in successfully bridging the gap to Silicon Valley. They achieve this by investing in talent through a 4M EUR fund, which sponsors four Entrepreneurs in Residence each year who temporarily join them in the Italian Alps.

Fabrizio has given talks across Europe sharing his journey as a European entrepreneur who spent 23 years in Silicon Valley before returning to Europe to contribute what he had learned. His presentation also highlights why Silicon Valley remains relevant, though it’s no longer essential for an entire company to be based there. And of course ends with the reasons why he started the Liquid Factory and why it makes sense to apply. These talks typically spark engaging Q&A sessions.

Seminar, Nobuko Yoshida, Monday 3rd February

Prof. Nobuko Yoshida of Oxford University is visiting us next Monday (3 February).

Nobuko has kindly agreed to give a talk on her work during her visit. The talk will be at 1pm in JC1.33a.

Nobuko Yoshida, University of Oxford, UK
https://www.cs.ox.ac.uk/people/nobuko.yoshida/

Title
Multiparty Session Types: Separation and Encodability Results

Abstract

Multiparty session types (MPST) are a type discipline for enforcing the structured, deadlock-free communication of concurrent and message-passing programs. Traditional MPST have a limited form of choice in which alternative communication possibilities are offered by a single participant and selected by another. Mixed choice multiparty session types (MCMP) extend the choice construct to include both selections and offers in the same choice. This talk first introduces the history and background of types for communications and multiparty session types, relating to the history of Computer Science in Oxford. This talk then presents a mixed choice synchronous multiparty session calculus and its typing system, which guarantees communication safety and deadlock-freedom. We then discuss the expressiveness of nine subcalculi of MCMP-calculus by examining their encodability (there exists a good encoding from one to another) and separation (there exists no good encoding from one calculus to another). The highlight is the binary (2-party) mixed sessions by Casal et al (2022) is strictly less expressive than the MCMP-calculus.

A joint work with Kirstin Peters appeared in LICS’24 (https://arxiv.org/abs/2405.08104)

About the speaker. Nobuko Yoshida is Christopher Strachey Chair of Computer Science in University of Oxford. She is an EPSRC Established Career Fellow and an Honorary Fellow at Glasgow University. Last 10 years, her main research interests are theories and applications of protocols specifications and verifications. She introduced multiparty session types [ POPL’08, JACM ] which received Most Influential POPL Paper Award in 2018 (judged by its influence over the last decade). This work enlarged the community and widened the scope of applications of session types, e.g. runtime monitoring based on Scribble (co-developed with Red Hat) has been deployed to other projects such as cyberinfrastructure in the US Ocean Observatories Initiative (OOI); and widened the scope of her research areas. She received the Test-of-time-award from PPDP’24 and the best paper awards from CC’20, COORDINATION’23 and DisCoTech’23. She received the third Suffrage Science Awards for Mathematics and Computing from MRC for her STEM activity. She is an editor of ACM Transactions on Programming Languages and Systems, ACM Formal Aspects of Computing, Mathematical Structures in Computer Science, Journal of Logical Algebraic Methods in Programming, and the chief editor of The Computer-aided Verification and Concurrency Column for EATCS Bulletin.

PGR Seminar with Zhongliang Guo

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

Below is a title and Abstract for Zhongliang’s talk– Please do come along if you are able.

Title: Adversarial Attack as a Defense: Preventing Unauthorized AI Generation in Computer Vision

Abstract: Adversarial attack is a technique that generate adversarial examples by adding imperceptible perturbations to clean images. These adversarial perturbations, though invisible to human eyes, can cause neural networks to produce incorrect outputs, making adversarial examples a significant security concern in deep learning. While previous research has primarily focused on designing powerful attacks to expose neural network vulnerabilities or using them as baselines for robustness evaluation, our work takes a novel perspective by leveraging adversarial examples to counter malicious uses of machine learning. In this seminar, I will present two of our recent works in this direction. First, I will introduce the Locally Adaptive Adversarial Color Attack (LAACA), which enables artists to protect their artwork from unauthorized neural style transfer by embedding imperceptible perturbations that significantly degrade the quality of style transfer results. Second, I will discuss our Posterior Collapse Attack (PCA), a grey-box attack method that disrupts unauthorized image editing based on Stable Diffusion by exploiting the common VAE structure in latent diffusion models. Our research demonstrates how adversarial examples, traditionally viewed as a security threat, can be repurposed as a proactive defense mechanism against the misuse of generative AI, contributing to the responsible development and deployment of these powerful technologies.

Distinguished Lecture series 2024

This years Distinguished Lecture series was delivered yesterday ( Tuesday 12th March) by Professor Neil Lawrence, University of Cambridge

In his talk on, ‘The Atomic Human Understanding Ourselves in the Age of AI’ he gave an overview of where we are now with machine learning solutions, and what challenges we face both in the near and far future. These include the practical application of existing algorithms in the face of the need to explain decision-making, mechanisms for improving the quality and availability of data and dealing with large unstructured datasets.