DEPLYOING mHEALTH QUESTIONNAIRES TO UNDERSTAND HOW COGNITIVE VITALITY EVOLVES

DERIVING PHENOTYPES OF PARTICIPANTS AND MONITORING THEIR EVOLUTION

ANALYSIS, MONITORING AND EVOLUTION

Multi-layer networks lend themselves for the construction of phenotypes of individuals, because intra-layer and inter-layer similarity can be used to express elaborate forms of proximity between individuals in the same layer and between features/layers. However, the features of these individuals change over time: people grow older, degenerative diseases lead to cognition decline, exercising leads gradually to better health. Translating this evolution into temporal learning in multi-layer networks is challenging: it is possible to specify snapshots at regular time intervals and build subgroups at each snapshot, but subgroups of different snapshots will not match with each other, although they will contain partially the same individuals. This leads to the problem of attempting to optimize subgroup quality at each snapshot versus enforcing subgroup continuity across snapshots. In this project, we propose a multi-layer snapshot network evolution approach that encompasses (a) a method for subgroup matching across snapshots, even for subgroups that do not have exactly the same set of layers, (b) a model of evolution that covers different forms of subgroup change, and (c) an evaluation mechanism that assesses the contribution of subgroups over time in predicting a target variable. We evaluate our approach for the task of predicting the performance of mHealth users in cognitive assessments.