Please note that this is not the official CLASS website. The official CLASS website can be found here.
CLASS - Cognitive-Level Annotation using Latent Statistical Structure - is a 3 year research project funded by the European Union. This webpage describes the contributions of the MPI Tübingen to the project.
Class will develop a basic cognitive ability for use in intelligent content analysis: the automatic discovery of content categories and attributes from unstructured content streams. The demonstrators will focus on object recognition and scene analysis in images and video with accompanying text streams. Autonomous learning will make recognition more adaptive and allow more general classes and much larger and more varied data sets to be handled.
| Matthew Blaschko |
| Peter Gehler |
| Christoph Lampert |
| Sebastian Nowozin |
| Bernhard Schölkopf |
| Gökhan BakIr - now at Google |
| Olivier Chapelle- now at Yahoo |
| Matthias Franz - now at HTWG Konstanz |
The following is a list of completed or ongoing projects done at the MPI. For other projects within CLASS, please see the official CLASS website.
|
Automatic Kernel Selecion. We present an algorithm (Infinite Kernel Learning,IKL) which is capable to learn linear combinations of kernels from very general classes. In fact these classes can be even infinite dimensional, e.g. all Gaussian kernels with p.d. covariance matrix. The algorithm solves a problem directly generalized from the Multiple Kernel Learning framework. (project details) |
|
Tuwo - C++ library for high level computer vision tasks. Includes feature extraction, supervised and unsupervised learning algorithms (project details) |
|
Efficient Subwindow Search (ESS). Object Localization by Globally Optimal Branch-and-Bound Search (C++), includes Python module for learning with structured regression. (source code) |
|
Structured Feature Spaces for Object Recognition Features extracted from digital images and videos are inherently interdependent. In 2D images, the dependencies include co-occurence and relative geometry, while in videos an ordering relation is added due to the temporal dimension. We explore learning in structured feature spaces that faithfully represent these dependencies. To this end, we model images as attributed graphs in which objects constitute subgraphs. Videos are modeled as sequences in which actions become subsequences. Objects and actions are recognized by identifying discriminative subgraphs and subsequences. (See gboost and pboost ) |
|
Multiple Instance Learning Multiple Instance Learning is a learning framework which to deal with ambiguous data. Label information is not available on a data point level but rather for a bag of possibly more than one data point. We are designing learning algorithms which are able to cope with this ambiguity. We are also exploring how tasks such as object recognition or named entity recognition can be cast as MIL problems and solved within this framework. ( project details ) |
|
Matlab Toolbox for Kernel Methods: The Spider. We are designing and developing a matlab toolbox for kernel methods. The goals of this project are: to build a general purpose kernel methods library including different induction principles such as (but not limited to) online/batch learning, active learning, etc., and to build a platform with different datasets and a code repository where researchers could exchange results and reproduce experiments.The spider is intended to be a complete object orientated environment for machine learning in Matlab. (project details) |