clampert - Posted on February 10, 2016 at 3:34 am.
|Our research lies at the interface between computer vision and machine learning. We solve computer vision problems using learning methods, and we develop learning techniques that are inspired by, but not limited to, problems that occur in computer vision. We aim for principled solutions based in theory rather than heuristic ones, always trying to understand the conceptual potential as well as limitations of the method we develop in addition to evaluating their practical usefulness.
Specific topics that we target are transfer learning (task learning, domain adaptation, lifelong learning, zero-shot learning), as well as structured prediction and learning (graph labeling, multi-label learning). We apply the techniques we develop to problems such as object recognition and localization in natural images, semantic segmentation, and interpretable representations (e.g. attributes).