DTAI Projects
BOF-DOC Relational Learning for Text and Image
Period: 09-2008 → 08-2011
Subgroup: ml
Type: project
Members:
Both text and images contain a lot of information that can be used to help interpret, complete and even reason about the image and the text. Whereas this is straightforward for humans, text and image-processing still are very challenging for machines. One of the reasons for this is that humans employ knowledge that is not (yet) available to machines. However, within artificial intelligence a lot of progress has been made on using and learning knowledge. Often relational representations are employed for this purpose. They do not only allow one to characterize properties of objects, but also to capture their relationships. This in turn provides essential information for interpreting text and images.
In this project, we shall investigate how relational representations and relational learning techniques can be employed to interpret text and image. To this aim, we shall investigate
- how relational representations can be generated from text,
- how relational representations can be generated from images,
- how relational representations for text and images can be integrated,
- how relational learning techniques can be used to improve the performance of text- and image processing systems.
Traffic analysis has been chosen as the application domain for evaluating the techniques and methods to be developed in this project. The reason is that this is interesting from both a scientific and an economic point of view, and that there exists quite some data about this domain that could be used in the project.
← return to the projects
