Konyushkova, KseniaSznitman, RaphaelFua, Pascal2018-01-242018-01-242018-01-242017https://infoscience.epfl.ch/handle/20.500.14299/144442In this paper, we suggest a novel data-driven approach to active learning (AL). The key idea is to train a regressor that predicts the expected error reduction for a candidate sample in a particular learning state. By formulating the query selection procedure as a regression problem we are not restricted to working with existing AL heuristics; instead, we learn strategies based on experience from previous AL outcomes. We show that a strategy can be learnt either from simple synthetic 2D datasets or from a subset of domain-specific data. Our method yields strategies that work well on real data from a wide range of domains.active learningmeta-learningtext::conference output::conference proceedings::conference paper