On Structured Output Training: Hard Cases and Efficient Alternatives
Speaker(s) : Thomas Gärtner (Fraunhofer IAIS)
State-of-the art structured-output learning algorithms have been applied in many important real world applications. However, there are also many structured-output problems in which these algorithms cannot be applied efficiently. In this talk I will give a simple example of such a problem, show why training the state-of-the-art algorithms on this problem is not feasible, and propose alternative algorithms. I will show that these algorithms can be trained efficiently under assumptions orthogonal to those made by the state-of-the-art.
Thomas.Baerecke (at) nulllip6.fr