Goal and Method
Our research is in the area of machine learning, with methods reaching into the field of mathematical optimization, and with applications in the fields of computer vision and biomedical image analysis.
The goal of our research is to construct machine learning algorithms for computer vision and biomedical image analysis that surpass the human in terms of accuracy and processing time. This goal is motivated by the broad impact such algorithms can have on the society, for instance, by accelerating scientific research in biology and medicine through intelligent microscopy, and by enabling the design of autonomous devices in engineering.
In pursuit of this goal, we construct and analyze mathematical abstractions of machine learning tasks in the form of optimization problems. We define, implement and apply algorithms for solving these problems exactly or approximately, and we compare the accuracy and processing time of these algorithms to that of humans with respect to practical metrics and data. This approach connects our research to a rich body of knowledge in the field of mathematical optimization on which we build.
Analysis and Optimization of Graph Decompositions by Lifted Multicuts
(focus on methods and mathematical foundations)
Joint Graph Decomposition & Node Labeling: Problem, Algorithms, Applications
(focus on practicality and application)