Research
Teaching computers to understand visual and other data has become an astonishing success story due to advances in machine learning and optimization and the arrival of powerful parallel computing devices together with massive amounts of visual data for training. In order to further extend the reach of machine learning and improve its performance further it is helpful to integrate classical algorithms into machine learning pipelines. It thus becomes possible to benefit from the best of both worlds, the data-driven nature of machine learning and the well understood nature of classical algorithms.
To this end Paul Swoboda develops innovative tools for extending the use of machine learning by combining them with other methodologies, mainly optimization. Optimization allows the explicit modeling of constraints on the output of a system and hence offers the possibility to get more inductive biases into machine learning systems. With these tools Paul Swoboda competes on a wide array of basic computer vision tasks, including segmentation, matching and tracking.
Research Interests
- Computer Vision
- Machine Learning
- (Combinatorial) Optimization
Courses
- Computer Vision
- Machine Learning