Jaques, N., Taylor, S., Nosakhare, E., Sano, A., Picard, R. In Proc. NIPS Workshop on ML in Health, Barcelona, Spain, December 2016. **BEST PAPER AWARD**

Abstract

Multi-task Learning (MTL) is applied to the problem of predicting next-day health, stress, and happiness using data from wearable sensors and smartphone logs. Three formulations of MTL are compared: i) Multi-task Multi-Kernel learning, which feeds information across tasks through kernel weights on feature types, ii) a Hierarchical Bayes model in which tasks share a common Dirichlet prior, and iii) Deep Neural Networks, which share several hidden layers but have final layers unique to each task. We show that by using MTL to leverage data from across the population while still customizing a model for each person, we can account for individual differences, and obtain state-of-the-art performance on this dataset.