Kristian Kersting - "Relations and probabilities: Friends, not foes"
AI has made tremendous progress so far. However, a lot remains to be done
if we are to reach the goals imagined by the early pioneers.
One of the bottlenecks is the traditional gap between logical and statistical AI:
Logical AI has mainly focused on complex representations,
and statistical AI on uncertainty. Intelligent agents, however, must be
able to handle the complexity and uncertainty of the real world.
Recent years have witnessed several successes in combining probability
and (subsets of) first-order logic in the field of statistical relational
learning. In this talk, I survey some of the success stories. Specifically, I
present non-parametric relational approaches to preference and reinforcement learning
as well as lifted inference algorithms for computing marginal probabilities
from relational probabilistic models. The latter inference approaches are lifted because
they work directly at the level of groups of atoms, eliminating all the
instantiations of a set of atoms in a single step, in some cases
independently of the number of these instantiations.
These contributions advance the theoretical understanding of statistical
learning with large models and continuous values. More importantly,
they let us dream of a next generation of AI.
Joint work with Babak Ahmadi, Sriraam Natarajan, Zhao Xu, Volker Tresp, Brian Milch,
Luke Zettlemoyer, Michael Haimes, Leslie Kaelbling, Kurt Driessens, Luc De Raedt, and
many more. The talk is partly based on:
- K. Kersting, B. Ahmadi, S. Natarajan. Counting Belief Propagation. In A. Ng, J. Bilmes, editors, Proceedings of the 25th Conference on Uncertainty in Artificial Intelligence (UAI 2009), Montreal, Canada, June 18-21 2009.
- B. Milch, L. Zettlemoyer, K. Kersting, M. Haimes, L. Pack Kaelbling. Lifted Probabilistic Inference with Counting Formulas. Proceedings of the Twenty-Third AAAI Conference on Artificial Intelligence (AAAI 2008), Chicago, Illinois, USA, July 13-–17 2008.
- K. Kersting, Z. Xu. Learning Preferences with Hidden Common Cause Relations. Proceedings of the European Conference on Machine Learning and Principles and Practice of Knowledge Discovery in Databases (ECML PKDD 2009), Bled, Slovenia, Sept. 7-11 2009.
- K. Kersting, K. Driessens. Non-Parametric Policy Gradients: A Unified Treatment of Propositional and Relational Domains. In A. McCallum, S. Roweis, editors, Proceedings of the 25th International Conference on Machine Learning (ICML 2008), Helsinki, Finland, July 5-9 2008
Kristian Kersting is the head of the "statisitcal relational activity mining" (STREAM) group
at Fraunhofer IAIS, Bonn, Germany, a research fellow of the University of Bonn, Germany, and
a research affiliate of the Massachusetts Institute of Technology (MIT), USA. He received his
Ph.D. from the University of Freiburg, Germany, in 2006. After a PostDoc at MIT, he joined Fraunhofer IAIS
in 2008 to build up the STREAM research group using an ATTRACT Fellowship. His main research interests are statistical relational learning, acting under uncertainty, and robotics.
He published over 40 peer-reviewed papers. He has received the ECML Best Student Paper Award in 2006
and the ECCAI Dissertation Award 2006 for the best European dissertation in the field of AI. He gave several tutorials on statistical relational learning and decision-theoretic
planning in relational domains at international conferences (AAAI, ECML/PKDD, ICAPS, IDA, ICML, ILP)
and co-chaired the international workshop on Mining and Learning with Graphs (MLG-07), the Dagstuhl
seminar 071611 on Statistical Relational Learning in 2007, and the international workshop on statistical
relational learning (SRL-09). He served as area chair for statistical learning for ECML in
2006 and 2007, and on the program committees of several international conferences and workshops such
as AAAI, ECAI, ECML/PKDD, ICML, IJCAI, ILP, KDD, RSS, and SRL. He is/was a guest co-editor of
special issues of the Annals of Mathematics and AI, the Journal of Machine Learning Research, and
the Machine Learning Journal. He serves on the editorial board of the Machine Learning Journal and the Journal of Artififical Intelligence Research.
Eyke Hüllermeier - "Learning Valued Preference Structures: Toward an Alternative Decision-Theoretic Framework for Machine Learning"
Bayesian decision theory is a commonly accepted framework in which problems of supervised learning can be formalized and analyzed in a convenient way. In this talk, we propose an alternative framework which is built upon the concept of a valued (fuzzy) preference structure and, thus, establishes a close connection between machine learning and fuzzy logic. A preference structure is a relational structure which, for each pair of decision alternatives, specifies a degree of strict preference, a degree of indifference, and a degree of incomparability. Comparing pairs of alternatives is quite natural and, from a machine learning point of view, consorts very well with the all-pairs binary decomposition scheme commonly used to reduce polychotomous to dichotomous classification problems. Based on related methods, the learning of models for predicting preference structures can be accomplished, which is interesting for several reasons. Notably, it makes existing techniques for decision making on the basis of fuzzy preference structures amenable to different types of prediction problems, including conventional classification but also generalizations thereof. Moreover, the distinction between indifference and incomparability between alternatives comes along with an interesting distinction between different types of uncertainty in predictive modeling and thereby supports sophisticated prediction and post-processing strategies.
Eyke Hüllermeier is with the Department of Mathematics and Computer Science at Marburg University (Germany), where he holds an appointment as a Full Professor and heads the Knowledge Engineering & Bioinformatics Lab. He holds M.Sc. degrees in mathematics and business computing, a Ph.D. in computer science, and a Habilitation degree, all from the University of Paderborn (Germany). His research interests are focused on machine learning and data mining, fuzzy set theory, uncertainty and approximate reasoning, and applications in bioinformatics. He has published numerous research papers on these topics in respective journals and major international conferences. He is a Member of the IEEE, the IEEE Computational Intelligence Society, and a board member of the European Society for Fuzzy Logic and Technology (EUSFLAT). He is on the editorial board of several journals, including Fuzzy Sets and Systems, Soft Computing, the International Journal of Uncertainty, Fuzziness and Knowledge-Based Systems, and the International Journal of Data Mining, Modeling and Management. Moreover, he is a co-ordinator of the EUSFLAT working group on Learning and Data Mining, and the head of the IEEE CIS Task Force on Machine Learning.