Research
REPRESENT / REASON / QUESTION
My research asks how AI systems can learn the structure of the world, reason about alternative explanations and outcomes, and make their decisions open to human scrutiny. Object-centric learning, causal reasoning and explainability provide complementary ways to approach these questions.
Objects & structure
Learning representations that reflect objects, their properties and the structure of a scene.
RepresentCauses & alternatives
Understanding interventions, counterfactuals and the mechanisms behind an observation.
ReasonExplanations & people
Building systems whose reasoning people can inspect, question and correct.
QuestionObjects and representations
I study representations that describe objects and their properties, with particular interest in grounding, identifiability and spatial ambiguity. This includes Grounded Object-Centric Learning, probabilistic slot attention, and identifiable object representations under spatial ambiguities.
Causes and counterfactuals
I am interested in the difference between generating a plausible alternative and modelling the consequences of an intervention. Related work includes diffusion counterfactual generation with semantic abduction, Factored Classifier-Free Guidance, and Bayesian causal discovery.
Explanations and human oversight
My work on explanations asks how people can understand and challenge a model’s decisions, including through argumentative exchanges. I want explanations to be part of how a system is used and questioned, rather than an afterthought attached to a prediction.
Research in practice
Medical imaging remains part of my research history and current collaborations. Maritime intelligence and fisheries monitoring provide a further setting in which robustness, incomplete observations and human oversight matter. My Maritime Impact articles will document that connection without treating a technical metric as proof of real-world impact.
Recent and earlier publications / Projects / Posters and presentations