Timezone: »

Learning Active Learning from Data
Ksenia Konyushkova · Raphael Sznitman · Pascal Fua

Mon Dec 04 06:30 PM -- 10:30 PM (PST) @ Pacific Ballroom #1 #None

In 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.

Author Information

Ksenia Konyushkova (EPFL)

I am Ksenia, a Ph.D. student in the CVLab at EPFL. In my research I apply methods from machine learning (and in particular active learning) to challenging problems in computer vision. I joined CVlab in 2014 and since then I have been working with Prof. Pascal Fua and Prof. Raphael Sznitman. I obtained my M.Sc. degree in Algorithms and Machine Learning from University of Helsinki. During that time, I also worked as a research assistant in the CoSCo group at HIIT. Before, I studied in Russia at the Higher School of Economics in the Faculty of Business Informatics and Applied Mathematics.

Raphael Sznitman (University of Bern)
Pascal Fua (EPFL, Switzerland)

More from the Same Authors