Thomas Gärtner
 |
Machine Learning Research Unit (E 194-06) Information Systems Engineering Institute Faculty of Informatics TU Wien (Technical University of Vienna) Tel.: +43 158801 188315 E-Mail: thomas.gaertner ... tuwien.ac.at Web: http://thomasgaertner.org Postal: Gußhausstraße 27-29, 1040 Vienna, Austria |
- I joined the TU Wien in October 2019 as a Professor of Machine Learning. We are hiring!!!
- Since 2019, I am a member of the Steering Committee of ECML PKDD and of the competence center for machine learning Rhine-Ruhr ML2R.
- We published a paper on Scalable Learning in Reproducing Kernel Kreı̆n Spaces at ICML'19 (A*, top 4%; joint work with Dino Oglic).
- I was Program Co-chair for ECML PKDD 2018, the largest European conference on Machine Learning and Data Mining (A, top 18%; with colleagues from IBM Research, ISI Foundation, and University College Dublin). Proceedings will be available online for free after the conference as per kind agreement with Springer.
- We published a paper on Learning in Reproducing Kernel Kreı̆n Spaces at ICML'18 (A*, top 4%; joint work with Dino Oglic).
- Our submission on Active Search for Computer-Aided Drug Design to the Molecular Informatics special issue on Generative Models for Chemical Biology and Drug Design has been accepted and is about to be printed (joint work with colleagues and friends from UoN School of Chemistry, University of Bonn, Washington University in St. Louis, and GSK)
- We published a paper on Effective Parallelisation for Machine Learning at NIPS'17 (A*, top 4%), the preprint and a teaser video are online; the post-proceedings are still in the making (joint work with colleagues and friends from Fraunhofer IAIS, University of Bonn, and GOOGLE)
- We published a paper on Co-regularised support vector regression at ECML PKDD'17 (A, top 18%), the proceedings of which are now officially online. A preliminary version of that paper received the Best Paper Award from the Data Mining in Biomedical Informatics and Healthcare workshop'16 and was published in the ICDM workshop proceedings (joint work with colleagues and friends from University of Bonn, Fraunhofer IAIS, and Bonn-Aachen International Center for Information Technology)
- We published a paper on Active Search in Intensionally Specified Structured Spaces at AAAI'17 (A*, top 4%; joint work with colleagues and friends from University of Bonn and Washington University in St. Louis)
- We published a paper on Nyström Method with Kernel K-means++ Samples as Landmarks at ICML'17 (A*, top 4%; joint work with Dino Oglic)
Research Interests
My
main research interests are efficient and effective machine learning
and data mining algorithms. Machine learning considers the problem of
extracting useful functional or probabilistic dependencies from a sample
of data. Such dependencies can then, for instance, be used to predict
properties of partially observed data. Data mining is often used in a
broader sense and includes several different computational problems, for
instance, finding regularities or patterns in data. By efficiency I mean
on the one hand the classical computational complexity of decision,
enumeration, etc. problems but on the other hand also a satisfactory
response time that allows for effectiveness. By effectiveness I mean how
well an algorithm helps to solve a real world problem. My recent focus is on challenges relevant to the constructive machine learning setting where the task is to find domain instances with desired properties and the mapping between instances and their properties is only partially accessible. This includes structured output prediction, active learning/search, online learning/optimisation, knowledge-based learning and related areas. I am most interested in cases of this setting where at least one of the involved spaces is not a Euclidean space but for instance the set of graphs. My approach in many cases is based on kernel methods where I have focussed originally on kernels for structured data, moved to semi-supervised / transductive learning, and am currently looking at parallel/distributed approaches as well as fast approximations. The most recent knowledge-based kernel method was for instance focussing on interactive visualisations for data exploration. Application areas which I am often considering when looking for novel machine learning challenges are chemoinformatics and computer games.