Personalized Digital Health Modeling with Adaptive Support Users
A novel approach to personalized digital health modeling employs adaptively weighted support users, encompassing both alike and unlike individuals, to address the challenges posed by limited and unreliable user-specific data. This technique combines personal loss, similarity-weighted transfer, and contrastive regularization to mitigate false correlations. An iterative optimization algorithm is utilized to concurrently refine model parameters and user similarity weights. Testing across six tasks within four actual digital health datasets reveals steady enhancements compared to both population and personalized benchmarks.
Key facts
- Personalized models are essential in digital health due to individual heterogeneity.
- Personalization is limited by scarce and noisy user-specific data.
- Existing methods rely on population pretraining or data from similar users only.
- Proposed framework uses adaptively weighted support users including similar and dissimilar individuals.
- Objective integrates personal loss, similarity-weighted transfer, and contrastive regularization.
- Iterative optimization algorithm jointly updates model parameters and user similarity weights.
- Experiments on six tasks across four real-world digital health datasets show consistent improvements.
- Method outperforms population and personalized baselines.
Entities
Institutions
- arXiv