Existing techniques for machine learning from numerical data often give predictions that may appear to be numerically quite accurate, but are on the other hand obviously incorrect qualitatively. Consider for example the learning from data about the process of emptying of a container through the open drain. Such learning techniques typically predict that the water level will be decreasing, but then strangely at some moment, water level will increase again a little. The resulting numerical error may be small, but qualitatively such predictions are obviously unacceptable. In this talk, an approach to machine learning from numerical data will be presented that combines both qualitative and numerical learning, and ensures that the resulting predictions are qualitatively correct. We call this approach Q2 learning, which stands for Qualitatively faithful Quantitative learning. Induced numerical models are “qualitatively faithful” in the sense that they respect qualitative trends in the learning data. An advantage of Q2 learning is that the induced qualitative model enables a (possibly causal) explanation of relations among the variables in the modelled system. Moreover, as we show experimentally the qualitative model’s guidance of the quantitative modelling process leads to predictions that may be considerably more accurate than those obtained by state-of-the-art numerical learning methods. The experiments include applications to the modelling of environmental processes and an industrially relevant mechanical system.
Ivan Bratko is professor of computer science at University of Ljubljana, Faculty of Computer and Information Science. Professor Bratko has also directed the AI group at J. Stefan Institute in Ljubljana, and has been visiting professor at various universities in Europe and elsewhere. Professor Bratko has conducted research in machine learning, knowledge-based systems, qualitative modelling, intelligent robotics, heuristic programming and computer chess. His main interests in machine learning have been in learning from noisy data, combining learning and qualitative reasoning, constructive induction, Inductive Logic Programming and various applications of machine learning, including medicine and control of dynamic systems. Best known among Bratko's many publications are the books Prolog Programming for Artificial Intelligence (Addison-Wesley/Pearson Education, third edition, 2001), KARDIO: a Study in Deep and Qualitative Knowledge for Expert Systems (MIT Press, 1989; co-authored by I. Mozetic and N. Lavrac), and Machine Learning and Data Mining: Methods and Applications (Wiley, 1998; co-edited by R.S. Michalski and M. Kubat).
Sprecher: Dr. Ivan Bratko Wann: Mi, 4. Mai 2005, 12:30 Uhr (s.t.) Wo: E 2.69, Universität Klagenfurt