PGR Seminar with Mirza Hossain

The next PGR seminar is taking place this Friday 14th March at 2PM in JC 1.33a

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

Title: BioFuse: Optimizing Biomedical Embeddings with Foundation Models

Abstract: Pre-trained foundation models have revolutionized biomedical AI, excelling in specialized domains like radiology and histopathology. However, integrating multiple models remains a challenge due to compatibility and feature fusion issues. BioFuse is an open-source framework designed to optimize biomedical embeddings by automatically selecting and fusing the best model combinations. Leveraging 9 state-of-the-art foundation models and a grid search strategy, BioFuse generates task-specific embeddings that improve downstream classification. On the MedMNIST+ benchmark, it achieves SOTA AUC in 5/12 datasets while maintaining near-SOTA performance in others. Surprisingly, our experiments reveal strong cross-modal capabilities, where models trained on one modality perform well on others. With a high-level API and an extensible architecture, BioFuse streamlines model integration and paves the way for new insights in biomedical data fusion.

PGR Seminar with Ben Claydon and Erdem Kus

The next PGR seminar is taking place this Friday 28th February at 2PM in JC 1.33a

Below are Titles and Abstracts for Ben and Erdem’s talks – Please do come along if you are able.

Ben Claydon

Title: Mechanisms for Similarity Search

Abstract:

Similarity search encompasses the task of finding those objects in a large collection which are most alike to, in some way, an object presented by the user as a query. The domain of these objects is wide, from images to text to chemical structures. This task becomes yet harder when the database becomes extremely large, and a sublinear query time with respect to the database size becomes a requirement. This talk discusses why the problem becomes so hard when presented with complex data, and how algorithms and data structures can be engineered to serve these queries.

Erdem Kus

Title: Frugal Algorithm Selection

Abstract: When solving decision and optimisation problems, many competing algorithms (model and solver choices) have complementary strengths. Typically, there is no single algorithm that works well for all instances of a problem. Automated algorithm selection has been shown to work very well for choosing a suitable algorithm for a given instance. However, the cost of training can be prohibitively large due to running candidate algorithms on a representative set of training instances. In this work, we explore reducing this cost by choosing a subset of the training instances on which to train. We approach this problem in three ways: using active learning to decide based on prediction uncertainty, augmenting the algorithm predictors with a timeout predictor, and collecting training data using a progressively increasing timeout. We evaluate combinations of these approaches on six datasets from ASLib and present the reduction in labelling cost achieved by each option.

PGR Seminar with Sharon Pisani

The next PGR seminar is taking place this Friday 21st February at 2PM in JC 1.33a

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

Title: Digital Cultural Landscapes for Sustainable Development in Remote and Island Communities

Abstract: Heritage plays a crucial role in community identity and sustainable development, yet remote and island communities often face challenges in engaging with and protecting their landscapes. This research explores how emergent digital technologies—such as 3D modelling, VR, and AR—can enhance heritage engagement and contribute to sustainable development. Using a practice-led methodology, case studies from Scotland and Malta demonstrate how digital cultural landscapes can support climate action, institutional capacity-building, and sustainable communities. A sustainable virtual museum framework is being developed, linking heritage to real-world environmental and socio-economic challenges. This presentation highlights the findings from these case studies, and the next steps in developing an immersive digital environment for an underwater heritage site.

PGR Seminar with Sachin Yadav and Junyu Zhang

The next PGR seminar is taking place this Friday 14th February at 2PM in JC 1.33a

Below is a title and Abstract for Sachin and Junyu’s talks – Please do come along if you are able.

Sachin Yadav

Title: Reimagining the Digital Gig Economy: Evaluating the economic feasibility and technological capabilities of physical cooperative gig platform

Abstract: The gig economy, fuelled by digital platforms, has transformed the labour markets around the world, offering flexibility but often at the cost of security for the worker and fair compensation. This presentation explores platform cooperatives – a democratically owned and governed alternative – as a potential solution to these challenges. I will delve into the economic feasibility and technological capabilities of physical delivery cooperatives, comparing them to traditional investor-owned platforms. By examining key performance metrics, regulatory environments, and worker empowerment, my ongoing work will assess whether platform cooperatives can achieve a comparable level of service while fostering more equitable working conditions. This presentation aims to spark discussion on the future of the gig economy and the role cooperative models can play in creating a more sustainable digital labour landscape.

Junyu Zhang

Title: Engaging Culture Heritage with Authentic Characters to Support Inclusive Learning

Abstract: Digitalization opens up new opportunities for cultural heritage, and lately the exploration of virtual reality has created new forms of representation of cultural content for educational institutions, museum exhibitions, and heritage preservation organizations. High-fidelity technology allows virtual agents to simulate realistic human appearances and behaviour to interact and engage with their surroundings. This speech presents work-in-progress research regarding designing, creating and utilising authentic characters to strengthen the exhibition of cultural heritage. Through the discussion on research design and practice, this research examines the capability of characters to enrich immersion and communication with heritage. This presentation introduces the realism and authenticity of character design, clarifies the goals for digitalization for inclusive learning opportunities in SDG, and ends with future work.

PGR Seminar with Mustafa Abdelwahed and Maria Andrei

The next PGR seminar is taking place this Friday 6th December at 2PM in JC 1.33a

Below is a title and Abstract for Mustafa and Maria’s talks – Please do come along if you are able.

Mustafa Abdelwahed:

Title: Behaviour Planning: A toolbox for diverse planning

Abstract:

Diverse planning approaches are utilised in real-world applications like risk management, automated streamed data analysis, and malware detection. These approaches aim to create diverse plans through a two-phase process. The first phase generates plans, while the second selects a subset of plans based on a diversity model. A diversity model is a function that quantifies the diversity of a given set of plans based on a provided distance function.

Unfortunately, existing diverse planning approaches do not account for those models when generating plans and struggle to explain why any two plans are different.

Existing diverse planning approaches do not account for those models when generating plans, hence struggle to explain why any two plans are different, and are limited to classical planning.

To address such limitations, we introduce Behaviour Planning, a novel toolbox that creates diverse plans based on customisable diversity models and can explain why two plans are different concerning such models.

Maria Andrei:

Title: Leveraging Immersive Technology to Enhance Climate Communication, Education & Action

Abstract: Climate change represents one of the most pressing challenges of our time, not only in its environmental impacts, but also as a pivotal science communication problem. Despite widespread scientific consensus on the causes and mitigation strategies for climate change, public understanding remains deeply fragmented and polarized. This disconnect hinders the collective action required from individuals, organizations, and policymakers to combat global warming effectively. My research explores the potential of immersive technologies to bridge the gap between scientific knowledge and public understanding by leveraging experiential learning experiences to inspire the attitudinal and behavioural shifts necessary to address climate change.

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.

AI Seminar Wednesday 27th November – Lars Kotthoff

We have another exciting AI seminar coming up on Wednesday 27th November at 1pm.

This time our speaker is an alumnus!

When? 27/11/24, 1pm

Where? JCB 1.33B

Who? Lars Kotthoff

Lars Kotthoff is the Templeton Associate Professor of Computer Science, Founding Adjunct Faculty at the School of Computing, and a Presidential Faculty Fellow at the University of Wyoming. His research in foundational AI and Machine Learning as well as applications of AI in other areas (in particular Materials Science) has been widely published and recognized. Lars is a senior member of the Association for the Advancement of AI and the Association of Computing Machinery.

What?

Title: AI for Materials Science: Tuning Laser-Induced Graphene Production

Abstract: AI and machine learning have advanced the state of the art in many application domains. We present an application to materials science; in particular, we use surrogate models with Bayesian optimization for automated parameter tuning to optimize the fabrication of laser-induced graphene. This process allows to create thin conductive lines in thin layers of insulating material, enabling the development of next-generation nano-circuits. This is of interest for example for in-space manufacturing. We are able to achieve improvements of up to a factor of two compared to existing approaches in the literature and to what human experts are able to achieve, in a reproducible manner. Our implementation is based on the open-source mlr and mlrMBO frameworks and generalizes to other applications.

PGR Seminar with Carla Davesa Sureda

The next PGR seminar is taking place this Friday 22nd November at 2PM in JC 1.33a

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

Title:

Towards High-Level Modelling in Automated Planning

Abstract:

Planning is a fundamental activity, arising frequently in many contexts, from daily tasks to industrial processes. The planning task consists of selecting a sequence of actions to achieve a specified goal from specified initial conditions. The Planning Domain Definition Language (PDDL) is the leading language used in the field of automated planning to model planning problems. Previous work has highlighted the limitations of PDDL, particularly in terms of its expressivity. Our interest lies in facilitating the handling of complex problems and enhancing the overall capability of automated planning systems. Unified-Planning is a Python library offering high-level API to specify planning problems and to invoke automated planners. In this paper, we present an extension of the UP library aimed at enhancing its expressivity for high-level problem modelling. In particular, we have added an array type, an expression to count booleans, and the allowance for integer parameters in actions. We show how these facilities enable natural high-level models of three classical planning problems.

Doughnuts will be available! 🍩

AI Seminar Tuesday 19th November – Francesco Leofante

The School is hosting an AI seminar on Tuesday 19th November at 11am in JCB1.33A/B

Our speaker is Francesco Leofante from Imperial College London.

Title:

Robustness issues in algorithmic recourse.

Abstract:

Counterfactual explanations (CEs) are advocated as being ideally suited to providing algorithmic recourse for subjects affected by the predictions of machine learning models. While CEs can be beneficial to affected individuals, recent work has exposed severe issues related to the robustness of state-of-the-art methods for obtaining CEs. Since a lack of robustness may compromise the validity of CEs, techniques to mitigate this risk are in order. In this talk we will begin by introducing the problem of (lack of) robustness, discuss its implications and present some recent solutions we developed to compute CEs with robustness guarantees.

Bio:

Francesco is an Imperial College Research Fellow affiliated with the Centre for Explainable Artificial Intelligence at Imperial College London. His research focuses on safe and explainable AI, with special emphasis on counterfactual explanations and their robustness. Since 2022, he leads the project “ConTrust: Robust Contrastive Explanations for Deep Neural Networks”, a four-year effort devoted to the formal study of robustness issues arising in XAI. More details about Francesco and his research can be found at https://fraleo.github.io/.

PGR Seminar with Daniel Wyeth and Ferdia McKeogh

The next PGR seminar is taking place this Friday 15th November at 2PM in JC 1.33a

Below is a Title and Abstract for Daniel’s and Ferdia’s talks – Please do come along if you are able.

Daniel:

Deep Priors: Integrating Domain Knowledge into Deep Neural Networks

Deep neural networks represent the state of the art for learning complex functions purely from data. There are however problems, such as medical imaging, where data is limited, and effective training of such networks is difficult. Moreover, this requirement for large datasets represents a deficiency compared to human learning, which is able harness prior understanding to acquire new concepts with very few examples. My work looks at methods for integrating domain knowledge into deep neural networks to guide training so that fewer examples are required. In particular I explore probabilistic atlases and probabilistic graphical models as representations for this prior information, architectures which enable networks to use these, and the application of these to problems in medical image understanding.

Ferdia:

“Lessons Learned From Emulating Architectures”

Automatically generating fast emulators from formal architecture specifications avoids the error-prone and time-consuming effort of manually implementing an emulator. The key challenge is achieving high performance from correctness-focused specifications; extracting relevant functional semantics and performing aggressive optimisations. In this talk I will present my work thus far, and reflect on some of the unsuccessful paths of research.