David-Daniel Ojebiyi
Pronouns: He/Him
Academic and research departments
UKRI Centre for Doctoral Training in AI for Digital Media Inclusion, Surrey Institute for People-Centred Artificial Intelligence (PAI).About
My research project
Trustworthy and Aligned AIMy research focuses on multimodal AI alignment: developing systems whose generated outputs reliably reflect human instructions, intended meaning, and real-world structure.
I am particularly interested in unified multimodal models that combine language understanding, reasoning, and visual generation within a single architecture. My current work investigates reinforcement-learning and reward-modelling methods for improving structural and relational faithfulness in image generation. This includes using scene graphs, spatial relationships, and fine-grained semantic representations to evaluate and optimise whether generated images correctly capture the objects, attributes, and interactions described in a prompt.
More broadly, I am interested in scalable alignment methods that move beyond surface-level preference optimisation toward rewards grounded in compositionality, causality, and multimodal consistency. My long-term goal is to contribute to AI systems that are not only capable, but also interpretable, trustworthy, inclusive, and reliably aligned with the intentions of the people who use them.
Supervisors
My research focuses on multimodal AI alignment: developing systems whose generated outputs reliably reflect human instructions, intended meaning, and real-world structure.
I am particularly interested in unified multimodal models that combine language understanding, reasoning, and visual generation within a single architecture. My current work investigates reinforcement-learning and reward-modelling methods for improving structural and relational faithfulness in image generation. This includes using scene graphs, spatial relationships, and fine-grained semantic representations to evaluate and optimise whether generated images correctly capture the objects, attributes, and interactions described in a prompt.
More broadly, I am interested in scalable alignment methods that move beyond surface-level preference optimisation toward rewards grounded in compositionality, causality, and multimodal consistency. My long-term goal is to contribute to AI systems that are not only capable, but also interpretable, trustworthy, inclusive, and reliably aligned with the intentions of the people who use them.