Seminar: Multilabel classification with ensembles of random graphs

SpeakerJuho Rousu
AffiliationAalto University, Finland
DateFriday, 25 Oct 2013
Time10:00 - 11:00
LocationRoom 1.02, Malet Place Engineering Building
Event seriesComputer Science Seminars

We present new methods for multilabel classification, relying on ensemble learning with discriminative graph labeling classifiers, trained on random graphs imposed on the multilabels. We study different methods of forming the ensemble prediction, majority voting as well as two methods making use of marginals of the graphs, namely the maximum average marginal (MAM) and the average max-marginal methods (AMM). We compare the methods against the state-of-the-art machine learning approaches on a set of heterogeneous multilabel benchmark problems, and show that random graph ensembles are very competitive and robust, ranking first or second on most of the datasets. Overall, our results show that random graph ensembles are viable alternatives to flat multilabel and multitask learners.

The presentation is based on joint work with Hongyu Su that will be published in the proceedings of the 5th Asian Conference on Machine Learning (ACML2013), held on 13-15 November 2013, in Canberra, Australia.

iCalendar csml_id_167.ics