Study in arXiv: Temporal Representation for Affective Computing
A recent preprint available on arXiv (2608.07518) introduces a temporal representation method aimed at resolving the cadence mismatch observed in longitudinal studies of affective computing. This investigation utilizes nearly a year’s worth of data from 82 participants in the Providemus alz study and evaluates three types of wave-level mappings—levels, absolute drift, and proportional drift—to forecast 21 outcomes related to affect and cognition. Physiological, sleep, activity, and environmental signals are condensed into concise wave-level descriptors, which are analyzed using four regressors through leave-one-subject-out and leave-one-wave-out methodologies. Findings suggest that wave-to-wave absolute drift is most effective for predicting affective states, while cognitive outcomes may benefit from alternative mappings. The research underscores the significance of representation in managing sparse labels in longitudinal wearable sensing.
Key facts
- Preprint on arXiv with ID 2608.07518
- Uses data from 82 adults in the Providemus alz study
- Data spans almost a year
- Models 21 affect and cognition outcomes
- Compares three wave-level mappings: levels, absolute drift, proportional drift
- Uses four regressors
- Evaluation axes: leave-one-subject-out and leave-one-wave-out
- Performance reported as scaled MAE with mean and median across folds
Entities
Institutions
- arXiv