Latest events

AIML ��volg Seminar: Understanding Scene Understanding

Professor Minh Hoai Nguyen and Professor Greg Zelinsky

Date: Tue, 16 Dec 2025, 10:30 am - 11:30 am
Location: AIML

Abstract: How do humans represent a scene after a brief period of viewing? ��volg shows that the “gist” of a scene (i.e., the scene name and layout) is extracted almost immediately from the blurred visual periphery, but little is known about how scene understanding evolves with viewing fixations. We studied this question using an integrated behavioural and computational approach with the aim of identifying scene metamers—scenes generated by a latent diffusion model that humans inaccurately believe were scenes they had just viewed (original scenes).

AIML Special Presentation: Active Perception and Reasoning in Open Worlds

Date: Tue, 9 Dec 2025, 10:30 am - 11:30 am
Location: AIML

Building intelligent systems that can perceive, reason, and act in open worlds remains a grand challenge. In this talk, Dr Shijie Li will share his journey toward active perception—from unifying 2D vision-language understanding to structured 3D reasoning and high-level foresight.

AI on the Ground - Dr Andrew Goodman - AI for healthcare in Australian Indigenous communities

Date: Fri, 21 Nov 2025, 10:30 am
Location: AIML Atrium

This presentation outlines a 24-month project that explored the relevance of AI for Australian Indigenous communities and identified concerns around its use and deployment in healthcare.

AIML Special Presentation: Professor Marcus Hutter

Date: Wed, 12 Nov 2025, 1:30 pm - 2:30 pm
Location: AIML

There is significant interest in understanding and constructing generally intelligent systems that approach and ultimately exceed human intelligence. Universal AI is presented as a mathematical theory of machine super-intelligence. More precisely, AIXI is described as an elegant, parameter-free theory of an optimal reinforcement learning agent embedded in an arbitrary unknown environment, possessing essentially all aspects of rational intelligence. The theory reduces all conceptual AI problems to pure computational questions. After a discussion of its philosophical, mathematical, and computational ingredients, a formal definition and measure of intelligence are introduced, maximized by AIXI. AIXI is framed as the most powerful Bayes-optimal sequential decision maker, and general optimality results are outlined.

Special Guest Presentation: Reduction Mappings for Guided Neural Collapse

Date: Mon, 10 Nov 2025, 2:30 pm
Location: AIML Atrium

Many high-dimensional optimization problems exhibit rich geometric structures in their set of minimizers, often forming smooth manifolds due to over-parameterization or symmetries. When this structure is known it can be exploited through reduction mappings that reparametrize part of the parameter space to lie on the solution manifold. It can be shown that well-designed reduction mappings improve curvature properties of the objective, leading to better-conditioned problems and faster convergence rates for gradient-based methods. We demonstrate this effect in guiding a neural network towards neural collapse—a known optimal configuration for over-parameterized classifier models. Material in this talk is joint work with Evan Markou and Thalaiyasingam Ajanthan, and appears in publications at NeurIPS 2024 and 2025.

AIML Special Presentation: Historical reasoning and machine learning

Date: Fri, 7 Nov 2025, 3:00 pm - 4:30 pm
Location: AIML

How do historians learn from the past to identify patterns, make predictions, and improve their work? Are their heuristics helpful for further advancing the development of machine learning? In this workshop, Professor Marnie Hughes-Warrington explored examples of how prize-winning historians reason about the past, including through dynamic spatio-temporal scaling, the use of conditionals and counterfactuals, and in the citation of temporal markers to fix and give credence to information.

Responsible AI Summit for Future Education

Date: Thu, 6 Nov 2025, 10:00 am - 3:00 pm
Location: AIML

The Responsible AI ��volg (RAIR) Centre invited educators and researchers to the Responsible AI Summit for Future Education to explore how AI was transforming the way we teach and learn. The event offered practical insights for teachers and education professionals seeking to integrate AI responsibly into their classrooms and curricula. Through a panel discussion and interactive workshops, participants gained strategies to enhance digital literacy, improve student engagement, and prepare learners for an increasingly AI-driven world.

AIML ��volg Seminar: How to engage effectively on LinkedIn

Date: Tue, 21 Oct 2025, 10:30 am - 11:15 am
Location: AIML

With over a billion users, LinkedIn is filled with people sharing their professional lives, achievements, and goals. But are you among the loudest voices on the platform?

AI on the Ground Seminar: AI for Paediatric Fracture Detection on Imaging

Date: Mon, 20 Oct 2025, 10:30 am - 11:15 am
Location: AIML

This talk explored the development and evaluation of artificial intelligence for detecting fractures in children on X-rays—one of the most common but misdiagnosed emergency care conditions. Dr. Shelmerdine reviewed current evidence from commercial AI tools, discussed findings from a large multi-reader study, and shared perspectives from children and families on trust, consent, and AI in healthcare. The session also touched on early work around cost-effectiveness and value for the NHS, asking the central question: could AI genuinely improve outcomes in paediatric fracture care while delivering value to health systems?

AIML ��volg Seminar: Intelligent Molecular Design

Date: Tue, 7 Oct 2025, 10:30 am - 11:15 am
Location: AIML

Abstract: Proteolysis-targeting chimeras (PROTACs) represent a powerful strategy for targeted protein degradation, yet their design is hindered by the complexity of linker generation and degradability assessment. Our recent research unifies PROTAC generation and degradability prediction. We leverage graph-based architectures to generate chemically valid linkers guided by protein–ligand context and integrates predictive modelling to evaluate degradation potential. Experiments demonstrate the framework’s ability to produce diverse, drug-like candidates with reliable degradability predictions.

��volg