Active semi-supervised learning for multi-target regression

Maira Farias Andrade Lira, Luisa Cavalcante, Celine Vens, Ricardo Prudencio, Felipe Kenji Nakano

Track: Type B (Encore Abstracts)

Keywords: Active learning, Semi-supervised learning, multi-target regression

Abstract:

Recent works have proposed the combination of active and
semi-supervised learning techniques to efficiently incorporate unlabeled
data. The so-called active semi-supervised learning (ASSL) investigates
methods to efficiently construct predictive models by employing either a
domain expert or model to (pseudo)-label data. To the best of our knowl-
edge, ASSL has not been applied in the context of multi-target regression,
a predictive task where multiple continuous targets must be predicted. In
this work, we propose MASSTER, Multi-target Active Semi-Supervised
Training for Regression, a novel ensemble method that identifies the
most relevant instance-target pairs based on the variance in their predic-
tions. Further, as its semi-supervised component, our method incorpo-
rates a variation of both self-learning (MASSTER-SL) and co-training
(MASSTER-CT). Experiments using 8 benchmark datasets reveal that
our method provides superior results in most of the cases when compared
to the current state-of-the-art.

Serve As Reviewer: ~Felipe_Kenji_Nakano1

Submission Number: 13