1: \begin{abstract}%   <- trailing '%' for backward compatibility of .sty file

2: Multi-task learning leverages shared information among data

3:  sets to improve the learning performance of individual tasks.

4:  The paper applies this framework for data where each task is a

5:  phase-shifted periodic time series.

6: %

7:  In particular, we develop a novel Bayesian nonparametric model

8:  capturing a mixture of Gaussian processes where each task

9:  is a sum of a group-specific function and a component capturing

10:  individual variation, in addition to each task being phase shifted.

11: %

12:  We develop an efficient \textsc{em} algorithm to learn the

13:  parameters of the model. As a special case we obtain the

14:  Gaussian mixture model and \textsc{em} algorithm for phased-shifted

15:  periodic time series.

16: %

17: Furthermore, we extend the proposed model by using a Dirichlet

18: Process prior and thereby leading to an infinite mixture model that

19: is capable of doing automatic model selection. A Variational Bayesian

20: approach is developed for inference in this model.

21: %

22:  Experiments in regression, classification and class discovery

23:  demonstrate the performance of the proposed models using both

24:  synthetic data and real-world time series data from

25:  astrophysics. Our methods are particularly useful when the time

26:  series are sparsely and non-synchronously sampled.

27: \end{abstract}

28: