Seminar on Decision Jungles

In July 2014 I took a seminar on current topics in computer vision and machine learning with Prof. Bastian Leibe. My topics were Decision Jungles as introduced by J. Shotton et al. at NIPS 2013.


In this paper, we present and discuss decision jungles as proposed by Shotton et al. at NIPS 2013. Decision jungles are ensembles of randomly trained decision DAGs (directed acyclic graphs). The concept is closely related to that of random forests as proposed by Breiman. For this reason, while discussing the topic, we compare it random forests in order to highlight the similarities as well as the differences. The main contribution of this paper is the derivation of an efficient implementation of the LSEARCH optimization algorithm. Using our resulting open source library, we perform several experiments in order to (1) investigate the influence different parameter settings have on the resulting classifier; (2) compare the performance of decision jungles and random forests for the task of multi class classification on a variety of data sets.