Smart task orderings for active online multitask learning
Date
2014-04-26Citation:
Pang, S., An, J., Zhao, J., Li, X., Ban, T., Inoue, D., and Sarrafzadeh, A. (2014). Smart Task Orderings for Active Online Multitask Learning. . Proceedings of SIAM International Conference on Data Mining(Ed.), Philadelphia, Pennysylvania, USA.Permanent link to Research Bank record:
https://hdl.handle.net/10652/2970Abstract
This paper promotes active oMTL (i.e., Online Multitask Learning with task selection) by proposing two smart task ordering approaches: QR-decomposition Ordering and Minimal-loss Ordering, in which the optimal sequence of tasks for oMTL is computed as the training data/tasks are being presented. Our experimental results on four real-world datasets show that the proposed task orderings outperform all existing task ordering approaches to active oMTL.