Department of Computer, Control and Management Engineering

Speaker: Andreas Karwath (University of Freiburg)

Title: Graphs, Sequences of Graphs, and their Alignment

Time and location: Friday, May 20, 2011, 11:00, Aula Magna

Abstract:
In his talk, A. Karwath will present a short introduction into relational machine learning, i.e learning from complex data structures with applications in the fields of chemoinformatics. Commonly, when calculating distances between graphs, or in the chemical domain from small molecules, one employs a fragment- (or fingerprint-) based approach to detect shared subgraphs of the two graphs. In his talk, A. Karwath will first present a real world application to learn complex models (hypothesis) from chemical structures, based on their string representation. Using this method, SMIREP, he will present the rediscovery of expert knowledge within the field of quantitative structure activity recognition (QSAR).

Moving on from this application he will present how sequences of graphs can be aligned based on a pre-defined distance metric with some application to bioinformatics. Furthermore, he will outline the drawbacks of using a predefined distance metric for more complex structures. Using this as a motivation, he will present an approach to learn distances between graphs under the assumption that graphs occur in a sequence, i.e. evolving world models, and at least two classes of these sequences are present.

And finally, he will give a short outlook of representing complex graph structures and their relations to each other by using visualization techniques within the chemical domain.

Bio:
Andreas Karwath is an academic researcher in the research group for Autonomous Intelligent Systems (AIS) at the university of Freiburg, Germany. He received his Ph.D. from the University of Wales in 2003 in the Computational Biology and Machine Learning Lab. Since 2004, he holds a post-doc position at the University of Freiburg. His research areas are machine learning and data-mining applied to real world problems, in particular in the fields of chemo- and bioinformatics. He is also PI in in the FP7 project OpenTox (www.opentox.org). Recently, he became interested dealing with numerical data and applications to activity recognition and predictions of neural brain activities.

Contact: Giorgio Grisetti ([email protected])