Essex team to present Hyper-DFS research at NeurIPS

University of Essex researchers will present work from the Human-Centred Artificial Intelligence Lab at Neural Information Processing Systems (NeurIPS) 2026 in Sydney, placing the team alongside leading universities and technology companies including Google, Microsoft and Meta. The accepted main-track paper focuses on dynamic feature selection, a technique used in AI systems to decide which information is most useful for a task.

The research identifies why existing dynamic feature selection methods do not always find the best solution and proposes Hyper-DFS, a hypernetwork-based method designed to make AI systems more efficient and adaptable. Essex says the method performs far better than existing systems when handling combinations of information it has not seen before.

Potential applications include medical diagnosis, where collecting information can cost time and money. Dr Javier Andreu-Perez called the acceptance a significant achievement for the group, while first author Dr Javier Fumanal-Idocin said hypernetworks can generate neural networks that adapt to individual situations when data is incomplete.