Research Methods Optimisation

The DEMON Network is the international network for the application of data science and AI to dementia. As such, the research methods optimisation group overarches all other working groups: it’s purpose is to develop novel and innovative AI methods to tackle the Network’s Grand Challenges.

Alongside working closely with other working groups, members of the research methods optimisation group will focus on developing solutions to current challenges facing the successful application of AI to dementia research.

Multi-cohort learning

Much of the data used for dementia research comes from individual cohorts. While there are already many examples of AI successfully being applied to a single cohort to, for example, classify current dementia status or predict future dementia risk, the results are only applicable to the cohort for which they were built: when models are validated on a different cohort performance decreases.

The research methods optimisation group will develop methods to learn from multiple cohorts. We envisage that a model that has learnt from multiple cohorts will be superior to a single cohort model and as such will be more widely applicable.

Multimodal learning

In dementia research there are many data modalities being analysed. These include, for example, neuroimaging scans, blood plasma, gene sequences and clinical data. The application of AI is leading to rapid advances in all areas of dementia research. However, in order to leverage all of the information available to us, models that learn from multiple modalities of data are required. The research methods optimisation group will therefore work to develop new ways of learning from and combining different data modalities for a single outcome.